Pick a model by task and trust only dated checks: rankings move weekly, prompt formats last longer, and the Sora API shut down on 24 September 2026, which OpenAI has confirmed.1
- Sora has ended. Veo 3.1 and Seedance are the practical replacements.
- Check for a negative field. FLUX.2, Runway, Seedance 2.5 and ChatGPT Image do not accept one.
- Leaderboards are a poor single signal. A model can fall down the rankings while it improves.
- Seedance 2.5 makes longer clips. It offers native 30-second single-pass generation with co-generated audio.
- On ChatGPT Image, read revised_prompt. OpenAI rewrites your prompt, so a mismatch may come from the rewrite.
Every price here was correct when this site last checked it, between 22 Sep and 23 Sep 2026. Prices change often and this site no longer updates them, so check the vendor’s own page before you rely on one: BytePlus ModelArk (Seedance) pricing · Runway Gen-4.5 credit costs · Black Forest Labs pricing · Pollo AI pricing.
NEGATIVE: LINES ON THIS PAGEEach model entry below carries a suggested negative prompt. Four of the models on this page do not accept one. Black Forest Labs states that FLUX.2 "does not support negative prompts"; Runway states negatives are unsupported and that including one "may result in the opposite happening"; Seedance 2.5 has no negative field, so exclusions compete with the positive prompt in the same text; ChatGPT Image has no field either.
Where a field exists — Midjourney's --no, and Veo's negativePrompt on Google's Gemini Enterprise Agent Platform — keep the line short and specific. Kling recommends writing negatives inside the prompt, so on Kling treat the line as prompt text. Where it does not, read the line as a checklist for the positive prompt instead: not "no plastic skin" but "visible pores, uneven natural skin tone." Full per-vendor breakdown: what the vendors actually say.
One more from the vendor docs, for ChatGPT Image: OpenAI’s API automatically revises your prompt before generating, and returns the rewritten version in a revised_prompt field. When output does not match what you asked for, the cause may be the revision rather than your wording — so read revised_prompt before rewriting. OpenAI also states the model struggles with precise text placement and layout-sensitive composition, which prompt tuning does not fix.
--nonegativePrompt, on Google’s Gemini Enterprise Agent PlatformOpenAI has discontinued Sora. The consumer app and web product ended 26 April 2026. The API shut down on 24 September 2026, with no replacement: OpenAI’s Sora 2 model page now says the Sora 2 models and the Videos API “are no longer available”.341First-hand: until 13 Oct 2026 this page said the shutdown was scheduled and not confirmed; OpenAI’s model page now confirms it.
Anything still pointed at it has stopped working. The practical migration paths are Veo 3.1 if you were using it for realism and native audio — noting that each Veo clip is 4, 6 or 8 seconds and longer video means extending it, up to 148 seconds at 720p — or Seedance if you were using it for prompt adherence on commercial work. One catch for Veo: on the Gemini API, Google says the Veo 3.1 previews were shut down on 22 Oct 2026 and names Gemini Omni Flash (gemini-omni-1.1-flash) as the replacement, while Veo 3.1 stays generally available on Google Cloud’s Gemini Enterprise Agent Platform.567
Seedance 2.5 — previewed 23 Jun 2026 at ByteDance's FORCE conference, released 31 July 2026. Native 30-second single-pass generation with co-generated audio, and up to 50 multimodal references: 30 images, 10 video clips, 10 audio files — not 30 images alone.8 Resolution, updated 22 Sep 2026: ByteDance’s own API price list sells Seedance 2.5 at 480p, 720p and 1080p, and gives it no 4K row; Dreamina’s page still says “Create cinematic 4K videos with Seedance 2.5 in Dreamina.”, which this site has not confirmed.9
The video leaderboards, as read on 1 Oct 2026. Re-query them; do not trust this snapshot. Artificial Analysis now runs a new board, AA-Video-T2V v2.0, on a new scale (Kling 3.0 1080p (Pro) is fixed at 1000), so its Elo figures cannot be compared with earlier ones. Default view as fetched: Wan 3.0 (1157), Dreamina Seedance 2.5 (1143), MiniMax H3 768p (1138), FLUX 3 (1129), Gemini Omni Flash 1.1 (1120), Dreamina Seedance 2.0 (1118). Arena, text to video, updated 21 Sep 2026: gemini-omni-1.1-flash (1516), gemini-omni-flash (1513), flux-3-video (1493), grok-imagine-video-1.5-agent (1492), dreamina-seedance-2.0-720p (1479), wan3.0 (1476), dreamina-seedance-2.5-720p (1474), minimax-h3 (1460); its best Kling entry is kling-2.5-turbo-1080p, 28th, with no row for Kling 3.0 or 4.0.10
Runway Gen-4.5 confirmed out of the top 10. It led at launch in Dec 2025 with 1,247 Elo and was displaced by Seedance 2.0, HappyHorse-1.0 and the Kling/Veo cluster by mid-2026.11 On 22 Sep 2026 it sat 19th on the text-to-video board without audio, at 1214 Elo, with a rank range of 15 to 26.12 It also gained native audio generation and editing in May 2026, which closes its longest-standing gap against Veo — kept here because a model falling off a leaderboard while improving is exactly why leaderboards are a poor single signal.
How to read this page
Every entry below carries the date it was last checked, not the date it was written. Model rankings move weekly and several of these will be wrong within a quarter — which is why the date is on the card rather than in a footer.
Each entry carries the prompt format that works for that model and what it is actually good at.
These were written in mid-2026. Model capabilities move faster than any guide. The prompt formats are the durable part; the capability claims carry a date and will eventually be wrong.
Check the stamp before trusting a comparison.
Guides for the tools covered here. Each card includes a prompt format and a copyable template; settings and capability claims carry the date on the card.
The best generative AI models right now — with optimized prompt formats for each. Click any model to expand its full prompting guide.
Every card is a dated snapshot. The prompt formats last; the capability claims do not, and several will be wrong within a quarter, so check the stamp.
AI MODELS
- Current version: ChatGPT Images 2.5, released 8 September 2026; in the API,
gpt-image-2.5-sunburstandgpt-image-2.5-flare.15 - Selling Point: Best conversational image editing — refine through natural language in multi-turn dialogue
- Renders text inside images, with a limit OpenAI states itself: “Although significantly improved, the model can still struggle with precise text placement and clarity.”16
- 4-round iterative workflow produces commercial-grade output
- Understands complex multi-element scenes
- What it is: Google's fast, cheap image model — “engineered for velocity and scale where speed and cost are the primary operational constraints”.17
- Purpose-built for rapid iteration and prompt testing before committing to slower, higher-quality models
- Not for reference-heavy work. Google: “Not optimized for multiple reference inputs or multi-turn sequential editing.” It also tops out at 1K: “Gemini 3.1 Flash Lite Image only supports 1K resolution.”
- Ideal workflow: prototype here, then finish in Nano Banana 2.1 (up to 4K output, better text) or another model.18
- Best for: rapid concept testing, volume social content, quick client mock-ups
- What it is: a video model. Artificial Analysis lists HappyHorse among its video models, under Alibaba-ATH, and it appears in the text-to-video standings in the leaderboard note at the top of this page.19
- No prompting guide for it has been read at source for this page, so no prompt format is given here.
- Selling Point: Human motion for fashion, lifestyle and commercial clips (this site’s use, not a benchmark result)
- This card’s template puts [MOTION] first; ByteDance does not require it
- Identity lock required: "same consistent face, no morphing" prevents character drift
- Clips of 4 to 15 seconds through the API (Seedance 2.5 goes to 30), with a cost comparison here.20
- Best for: AI influencer video, cinematic human motion, brand lifestyle clips
- Selling Point: Best physics-accurate motion — water, fire, fabric, and impact sequences look physically real
- Kling 3.0 (2026): extended clips, improved facial consistency, strong image-to-video
- Kling 3.0 clips run from 3 to 15 seconds. Kling: “The new model generates up to 15 seconds of continuous video, with a flexible duration ranging from 3 to 15 seconds.”21 Kling 4.0 goes up to 30 seconds in one pass, Kling says.22 Kling says 4.0 rolled out to all users on 15 Oct 2026.23
- Best for: physics action, product reveal sequences, nature and environment video
- Selling Point: Best-in-class for brand commercial VFX and style transfer sequences
- Clips run 2 to 10 seconds, at 12 credits a second. Runway’s Gen-4.5 page lists “Supported durations 2 - 10 seconds”.24First-hand: until 22 Sep 2026 this line said “Clips are 5 or 10 seconds”, quoting Runway’s Gen-4 guide: “Gen-4 creates videos in 5 or 10 second durations”. That is Gen-4’s range, not Gen-4.5’s. An earlier version of this card was titled Gen-4 Turbo and said clips run up to 16 seconds.
- Strong at transformations, morphs, and surrealist visual effects for commercial use
- Best for: brand commercials, VFX sequences, style transfer, film-grade video production
- Selling Point: Highest prompt adherence of any image model — what you write is what you get, precisely
- Black Forest Labs lists four FLUX.2 tiers: “FLUX.2 [pro], FLUX.2 [flex], FLUX.2 [dev], and FLUX.2 [klein]”. An earlier version of this card described a “FLUX.2 Pro Ultra” tier, which Black Forest Labs does not list.25
- Settings: this site found no Black Forest Labs source for CFG, steps or sampler on FLUX.2 [pro], so none are given here
- Black Forest Labs now makes video too. FLUX 3, in preview since 4 Aug 2026, is its first video model: “one model trained across image, video, and audio, generating video with synchronized sound”. Text- or image-to-video costs $0.17 a second at hd, $0.29 at fhd, $0.40 at qhd and $0.80 at uhd, or $0.06 a second for a draft.26 Since 10 Sep 2026 one request can return a 2K or 4K clip, with no separate upscaling step.27
- Best for: technical compositions, detailed architecture, precise product visualization, ComfyUI pipelines
- Selling Point: Strongest editorial aesthetic and artistic composition of any image model — distinctly recognizable visual style
- Current default: V8.2, which Midjourney says “released as the default version on July 24, 2026.” It describes V8.2 as “an update focused on aesthetics, image quality, and Personalization.”28
- Use --ar for aspect ratio, --no instead of negative prompts, --q 2 for quality
- Best for: editorial fashion, creative direction, concept art, artistic campaigns
- The catch: you cannot take a track out. Since its partnership with Universal Music Group, announced 29 Oct 2025, Udio says “downloading of audio, video, and stems has been disabled”.29
- Udio v1.5 is a 2024 model. Udio’s post introducing it, with stem downloads as a new feature, was “Published on Jul 23, 2024”.30First-hand: until 22 Sep 2026 this card gave Udio’s selling point as “Stem separation and export”, dated v1.5 to 2026 and recommended it for “stems for mixing”. Stem downloads had been switched off before the card was checked on 22 Aug 2026.
- Best for: sketching and remixing songs inside Udio, not for a track you need to use elsewhere
- Suno v6 (9 Sep 2026) — Suno’s flagship, alongside v6-wild, “less predictable and more varied”, and v6-mini, “available to everyone”. “All models prior to v6 have been retired.” An earlier version of this card described v5.5 as current.31
- Full songs with lyrics, vocals, and production from a single text prompt — any genre, any mood
- Suno Studio (2026): multi-track editing, stem separation, and remix tools built into the platform
- Best for: AI film scores, brand content, social media, influencer soundtracks, full song production
[Verse], [Chorus], [Bridge]). Putting lyrics in the style box, or genre in the lyrics box, is the most common reason output ignores half the prompt.Suno writes its own style example in sentences: “Style: Contemporary R&B song at 92 BPM in F minor.” Be specific — Suno says to “mention genre, mood, keywords, and instrumentation”.
What you do not want goes in Exclude, in Custom Mode under Advanced Options.32
- Seedream 5.0 arrived in February 2026; 5.0 Pro on 8 July 2026. ByteDance's image model, and the sibling of Seedance — same platform, same documentation style.
- Text rendering is the headline. Small fonts render more accurately, character repetition is reduced, and the bias toward bold is corrected.
- Up to six reference images, merged into one result.
- 5.0 edits by partial selection and pen input; 5.0 Pro adds point and lasso selection and layer separation.33
@Image1 tagging.- There is no prompt box. On text-to-speech, the text you submit is the prompt — punctuation, capitalisation and tags are your only controls over delivery.
- New: Eleven v4 and v4 Turbo, launched 28 Sep 2026. ElevenLabs says v4 reads tone, pacing and emotion from context, works in more than 90 languages, and still takes inline audio tags such as
[excited, happy],[laughs]and even sound cues like[light rain]. v4 Turbo is the faster version, meant for live voice agents.35 - Not yet documented for v4: whether every v3 tag carries over, whether SSML break tags work, whether Eleven v3 stays available, and the v4 model IDs. The launch post does not say. The v3 and v2 rules below are marked as such.
- Eleven v3 and earlier models take different syntax. This is the single most important thing on this entry.
- Voice cloning holds identity across a content series; the voice you clone constrains what any tag can do to it.
[whispers], [excited], [sarcastic], [laughing], [sigh]. ElevenLabs: “Eleven v3 does not support SSML break tags.” Use punctuation and audio tags for pauses instead.<break time="1.5s" />, up to a 3 second maximum. ElevenLabs warns that too many break tags in one generation "can cause instability" — the voice may speed up or introduce artefacts. Use a few deliberately, not one after every line.It was a VERY long day [sigh] … nobody listens anymore. These work without any tag syntax and are the safest lever across versions.[shout] on a voice trained on whispering will not produce a shout. Tag effectiveness depends on the voice and its training samples, so a tag that works on one voice may do nothing on another. They also note Professional Voice Clones are not fully optimised for v3. Test your tags on your actual voice before scripting around them.36- Selling Point: Upload any long-form video and it automatically finds the most viral moments, reformats for vertical, and exports short clips with captions
- AI identifies hooks, engagement peaks, and viral patterns automatically — no manual editing
- Outputs TikTok, Reels, and Shorts-ready clips with bold captions in one pass
- Best for: podcasters, YouTubers, educators, and brands turning long content into daily social posts
- Selling Point: Full end-to-end faceless video automation — it writes the script, generates voiceover, creates visuals, adds captions, and auto-posts to TikTok, YouTube, Instagram, and Facebook daily.. Set it up once and it runs without you.
- Auto-posts to all 4 platforms simultaneously — with platform-optimized titles and hashtags for each
- Generates thumbnails, captions, and optimal posting times automatically — no Canva, no scheduling apps
- Best for: faceless YouTube channels, TikTok automation, and high-volume content creators
AI STUDIOS
An AI Studio is an all-in-one creative platform that bundles multiple AI capabilities — image generation, video, voice, editing, and automation — into a single interface. Unlike standalone models, studios are designed for full production workflows.
★ There are many AI studios in the market — it comes down to where you feel comfortable and what suits your workflow. Try a few and commit to the one that fits.
- Selling Point: Not a model — a full cinematic AI production studio
- Pioneering platform for Hollywood-grade AI video production
- AI aggregator: access multiple video models in one interface
- Upload a face photo → generate that person in any scene
- New, 1 Sep 2026: Genjutsu, a tool that swaps one thing in an existing video (an object, a place, a product or a character) and leaves the rest as it was37
- Best for: AI influencers, character films, brand content
- Selling Point: Hands-on AI influencer customization — build and control every detail of your AI persona with full creative control, from appearance to identity consistency across unlimited content
- Virtual try-on, branded outfit swaps, and lifestyle scenario generation — no photography needed
- Best for: AI influencer creation, brand ambassador content, fashion and beauty campaigns
- Selling Point: Upload a reference image and Glam AI handles the prompting automatically — template-driven beauty and fashion content without manual prompt writing
- Sponsored post creation and brand partnership content generation built into the workflow
- Best for: beauty brands, fashion influencers, product launch campaigns, e-commerce lookbooks
- Selling Point: Give it a URL, PDF, or prompt — it researches the topic and auto-generates a complete, polished multi-scene explainer or promotional video from scratch
- No timeline editing, no assets needed — fully autonomous video production from text input
- Best for: explainer videos, thought leadership content, product demos, educational content
- Selling Point: Converts any text, blog post, or script into a fully edited video with voiceover, music, captions, and stock footage — no editing skills needed
- 16M+ stock clips, automated scene matching, brand kit, and direct social media publishing
- Best for: marketers, YouTubers, agencies producing high-volume video content at scale
- Selling Point: 20+ AI image models in one platform — run the same prompt through FLUX, SDXL, Stable Diffusion variants simultaneously for comparison
- ControlNet, LoRA support, and inpainting built-in — advanced workflows without ComfyUI
- Best for: prompt engineers testing across models, designers exploring multiple visual aesthetics
- Selling Point: Visual workflow builder lets you chain multiple AI operations together (generate → edit → upscale → stylize) without any coding
- Strong AI character consistency and avatar creation — ideal for influencer content pipelines
- Best for: creators building repeatable AI workflows, AI influencer persona development
- Selling Point: Cinematic-quality text-to-video and image-to-video with a free plan — accessible entry point for creators new to AI video
- Strong realistic human motion and facial consistency, competitive with Kling and Runway
- Best for: cinematic short clips, AI influencer video, product showcases, social content
NEWSLETTER EXCLUSIVEFuture exclusive prompt packs and AI production workflows sent directly to subscribers. Subscribe free ↗
The Google suite, in one place
Google now ships roughly a dozen separate AI tools, and the confusing part is that several overlap. Grouped by what they actually do rather than by launch order:
One vendor, many products, no obvious map. Three of these do coding, two do video, two do design. Listing them is not endorsing them — it is making the overlap visible so you can pick one and ignore the rest.
Coding and agents
- What it does: handles coding tasks in the background, explains its changes, fixes bugs, prepares code for review
- The asynchronous pattern — you do not sit and watch it
- Review still required; see diagnosing a run
- What it does: a coding workspace where you assign tasks to agents across editor, terminal and browser
- Broader permission surface than a normal editor — three environments, not one
- Worth setting tiers before granting it much
- What it does: turns plain-language descriptions into small apps with editable visual workflows
- The visual workflow is the useful part — you can see what it built
- Closest thing here to vibe coding with a safety rail
Design and interface
- What it does: text prompts into editable UI designs and front-end code
- Idea to working interface without opening a design tool
- Generated front-end code still needs the tier-one checks
- What it does: collects ideas, generates visuals, remixes them into a direction
- Direction-finding rather than production
- Useful before you start prompting seriously, not after
Video and image
- What it does: realistic video with synchronised sound, dialogue and cinematic motion from text or an image
- 48kHz synchronised dialogue with sub-120ms lip-sync — as reported by reviewers. One July 2026 roundup calls it the only widely available model with it39
- But one clip is 4, 6 or 8 seconds. Longer needs extending: each extension adds 7 seconds, up to 148 seconds in total, at 720p only. Every Veo video carries Google’s hidden SynthID watermark.40First-hand: until 1 Oct 2026 this card was titled “Veo 3” and said “native clips top out at 8 seconds (some sources say 15)”. The “15” was wrong: Google’s own Veo page gives 4, 6 or 8 seconds per clip. It also accepts only ~3 reference images against Seedance's 50
- Where it runs is changing. Google says the Veo 3.1 previews on the Gemini API were shut down on 22 Oct 2026, with Gemini Omni Flash as the replacement; on Google Cloud’s Gemini Enterprise Agent Platform Veo 3.1 is generally available, with retirement “November 17, 2026 or later”. In the Gemini app, Google says Gemini Omni “will replace Veo”.56417
- The migration path if you were using Sora for realism and native audio
- What it does: builds scenes, manages characters, adjusts camera angles, maintains consistency across a story
- Two video models inside: Veo 3.1 and Gemini Omni Flash, which can also edit footage you bring. Flow also has Flow Agent, which plans and builds with you in conversation, and an Android app42
- The consistency claim is the interesting one — see character consistency
- Camera control assumes you know the vocabulary: camera grammar
- What changed: Imagen 4 was replaced by Google’s Nano Banana models. Google lists all three Imagen 4 models with a shutdown date of 17 August 2026, and now names
gemini-nano-banana-2.1, Nano Banana 2.1, as the replacement. Nano Banana 2 (gemini-3.1-flash-image), which Google named until October, is now deprecated, with no shutdown date announced.43 - New in October: Nano Banana 2.1, an update to Nano Banana 2, became generally available in the Gemini API on 6 October 2026.44
- Text rendering differs by model and still needs checking: see text inside images
- Comparison on the comparison page
Research and knowledge
- What it does: turns uploaded notes, PDFs, papers and transcripts into a workspace for questions and summaries
- Renamed Gemini Notebook in July 2026. Same product, existing notebooks intact, free tier continues45
- "Grounded" is the operative word — answers come from your sources rather than from training
- This is context engineering as a product
- What it does: analyses a business website to build its "Business DNA", then generates tailored campaigns and branded content
- Closest match to the creation-as-a-service model
- Whether a site is enough to infer a brand from is the open question
Several of these are experiments with real limits, and vendor lists rarely say so:
- Mixboard — US-only, waitlist
- Opal — US-first at launch, now reported in 160+ countries
- Stitch — regional beta. Originally Galileo AI, acquired by Google and rebranded
- Antigravity — free during preview, but free requests reported cut to roughly 20/day. It is a fork of VS Code
- Pomelli — beta, paid tiers signalled, reported in 170+ countries
Anything under Google Labs is explicitly an experiment. Whisk and ImageFX did not simply close: Google announced on 25 February 2026 that they “are moving directly into Flow”, and both old addresses now redirect to Flow. Free is not a permanent promise, and neither is existence — the Sora shutdown at the top of this page is the same lesson from a different vendor.46
This is one vendor's catalogue at one moment. Several of these overlap, and some may merge or be discontinued — as Sora was at OpenAI: the Sora app closed on 26 April 2026, and the API shut down on 24 September 2026.3471
Pick by what you need rather than by what exists. The decision ladder is on prompt, context, skill or fine-tune, and what it costs to run is on token economics.
Google's AI tools overlap, and several are experiments. Pick one by what you need, check where it is available, and do not treat free as permanent.
Related on this site: ComfyUI Explained
- 1OpenAI, Sora 2 model page, read at source 13 Oct 2026“The Sora 2 models and Videos API were shut down on September 24, 2026 and are no longer available.”; “There is no one-to-one replacement API.” ↑a“The Sora 2 models and Videos API were shut down on September 24, 2026 and are no longer available.”; “There is no one-to-one replacement API.” ↑b“The Sora 2 models and Videos API were shut down on September 24, 2026 and are no longer available.”; “There is no one-to-one replacement API.” ↑c“The Sora 2 models and Videos API were shut down on September 24, 2026 and are no longer available.”; “There is no one-to-one replacement API.” ↑d
- 2Vendor documentation: docs.bfl.ai · help.runwayml.com · Google Cloud Veo guide · Kling API docs. Checked 25 Aug 2026↑
- 3OpenAI, API deprecations page, read at source 1 Oct 2026“we notified developers using the Videos API and Sora 2 video generation model aliases and snapshots of their deprecation and removal from the API on September 24, 2026.” Replacement column: “---”. ↑ashutdown date “2026-09-24” for the Videos API,
sora-2andsora-2-pro, with no confirmation that it took place. ↑bstill lists the 24 September 2026 date with no note that the shutdown happened. ↑c - 4OpenAI Help Center, What to know about the Sora discontinuation and the API deprecations page, read at source 9 Sep 2026developers were notified 24 March 2026, the web and app experiences ended 26 April 2026, and all API video generation including Sora 2 and Sora 2 Pro stops on 24 September 2026. Also reported across Pixo, ChatCut, Kingy and Teamday, Apr–Jul 2026 ↑
- 5Google, Gemini API deprecations, last updated 12 Oct 2026, read at source 13 Oct 2026veo-3.1-generate-preview, veo-3.1-fast-generate-preview and veo-3.1-lite-generate-preview, shutdown “October 22, 2026”, recommended replacement “gemini-omni-1.1-flash”; the page notes that its dates “indicate the earliest possible dates on which a model might be retired”. ↑aveo-3.1-generate-preview, veo-3.1-fast-generate-preview and veo-3.1-lite-generate-preview, shutdown “October 22, 2026”, recommended replacement “gemini-omni-1.1-flash”; the page notes that its dates “indicate the earliest possible dates on which a model might be retired”. ↑b
- 6Google Cloud, Veo 3.1, Gemini Enterprise Agent Platform documentation, last updated 10 Oct 2026, read at source 13 Oct 2026veo-3.1-generate-001 and veo-3.1-fast-generate-001, “Launch stage: GA”, retirement date “November 17, 2026 or later”; veo-3.1-lite-generate-001, “Launch stage: Preview”. ↑aveo-3.1-generate-001 and veo-3.1-fast-generate-001, “Launch stage: GA”, retirement date “November 17, 2026 or later”; veo-3.1-lite-generate-001, “Launch stage: Preview”. ↑b
- 7Google, Video generation in the Gemini API, read at source 3 Nov 2026“Veo 3.1 Preview (veo-3.1-generate-preview, veo-3.1-fast-generate-preview, and veo-3.1-lite-generate-preview) in the Gemini API was shut down on October 22, 2026; migrate to Gemini Omni Flash (gemini-omni-1.1-flash) or the GA models available through the Gemini Enterprise Agent Platform to avoid service interruptions.” ↑a“Veo 3.1 Preview (veo-3.1-generate-preview, veo-3.1-fast-generate-preview, and veo-3.1-lite-generate-preview) in the Gemini API was shut down on October 22, 2026; migrate to Gemini Omni Flash (gemini-omni-1.1-flash) or the GA models available through the Gemini Enterprise Agent Platform to avoid service interruptions.” ↑b
- 8ByteDance’s own Seedance 2.5 page on Dreamina, read at source 11 Sep 2026“create 4K AI videos up to 30 seconds with up to 50 multimodal references”. Not re-verified: re-read in a browser on 17 Sep 2026, the page no longer carries that sentence; it now says “Generate 30-second videos from up to 50 references.” The split into 30 images, 10 video clips and 10 audio files is not on that page as re-read. Also verified 22 Aug 2026 against Artlist, Morphic, Wiro and VioEvo. ↑
- 9ByteDance, BytePlus ModelArk Pricing, page last updated 22 Sep 2026, read in a browser 22 Sep 2026dreamina-seedance-2-5-260628 “For 1080p outputs: Input without video: 11.7” (USD per million tokens) · Dreamina, Seedance 2.5, read at source 22 Sep 2026. First-hand: until 22 Sep 2026 this line said the resolution was disputed, “reported as 720p at launch, 1080p since”. ↑
- 10Artificial Analysis, AA-Video-T2V v2.0 leaderboard, no date on the page, read at source 1 Oct 2026Kling 3.0 1080p (Pro) “Anchor Fixed at 1000 as the reference point for this board’s Elo scale.” Arena, Text-to-Video leaderboard, last updated 21 Sep 2026, read at source 1 Oct 2026. The no-audio and image-to-video views were not re-read. Until 1 Oct 2026 this note gave the 22 Sep 2026 readings of the old board (Gemini Omni Flash 1233, Wan 3.0 1229 with audio) and said Seedance 2.5 had no row. ↑
- 11Runway, Introducing Runway Gen-4.5, 1 Dec 2025, read at source 23 Sep 2026“With 1,247 Elo points, Gen-4.5 currently holds the top position in the Artificial Analysis Text to Video benchmark”. Pinggy, Best Video Generation AI Models in 2026, updated 14 Jul 2026, read at source 23 Sep 2026: “Runway Gen-4.5, which led at launch in late 2025 with 1247 Elo, has dropped out of the top 10.” Also checked against an Artificial Analysis API reading via Banana Flow, 5 Aug 2026 (no link was recorded). ↑
- 12Artificial Analysis, Text to Video Leaderboard, no-audio view, read in a browser 22 Sep 2026“Runway Gen-4.5 1214”, rank range “15-26”. Re-query it; do not trust this snapshot. An earlier version of this line gave an Aug 2026 secondary reading of about 1,215 at rank 15. ↑
- 13Dreamina (ByteDance), Seedance 2.5, read at source 23 Sep 2026“You can create cinematic videos up to 30 seconds in standard mode or extend them to 180 seconds with the beta long-video mode.” ↑
- 14Pinggy, Best Video Generation AI Models in 2026, updated 14 Jul 2026, read at source 23 Sep 2026“Veo 3.1 still owns this with 48kHz speech generation.” ↑
- 15OpenAI, Introducing ChatGPT Images 2.5, read in a browser 16 Sep 2026; Image generation guide, read at source 16 Sep 2026. An earlier version of this card named ChatGPT Images 2.0 as current.↑
- 16OpenAI, Image generation guide, read at source 17 Sep 2026. An earlier version of this card said “Excellent text rendering” with no qualification.↑
- 17Google, Nano Banana image generation, read at source 13 Oct 2026“The fastest and cheapest Gemini image model, engineered for velocity and scale where speed and cost are the primary operational constraints.” An earlier version of this card called it “Nano Banana Pro 2”, a name Google does not use. ↑
- 18Google, Nano Banana image generation, read in a browser 13 Oct 2026Nano Banana 2.1 “maintains Flash-level speed and cost efficiency while delivering improved visual quality, text rendering, multi-turn consistency, and Google Search grounding across 1K, 2K, and 4K resolutions.” ↑
- 19Artificial Analysis, HappyHorse model family page, read in a browser 17 Sep 2026“Analysis of HappyHorse models and comparison to other video models across key metrics including quality, generation time, and price.” An earlier version of this card presented HappyHorse-1.0 as a text-to-image model, labelled it TOP-2 in April 2026, and gave it an image prompt template. None of that had a source, and it was removed. ↑
- 20BytePlus, Dreamina Seedance 2.5 tutorial, read at source 22 Sep 2026the duration row reads “4–30 seconds” for 2.5 and “4–15 seconds” for the 2.0 series. First-hand: until 23 Sep 2026 this line said “Clips up to 10 seconds at 24fps”. ↑
- 21Kling AI, Kling VIDEO 3.0 model user guide, read at source 17 Sep 2026. An earlier version of this card said up to 10 seconds.↑
- 22Kling AI, Kling 4.0 vs 3.0, 30 Sep 2026, read at source 1 Oct 2026“Kling 4.0 extends native generation to 30 seconds in a single pass.” (On 1 Oct the page said early access began 28 Sep and the full model was due in October.) ↑
- 23Kling AI, Kling 4.0 at Work: Bringing Creative Vision into Production, 27 Oct 2026, read at source 3 Nov 2026“Since Kling 4.0 rolled out to all users on October 15, creative teams across film, advertising, and animation have been putting these capabilities to work.” · Kuaishou Technology, press release on Kling 4.0, carried by Bernama, 19 Oct 2026, read at source 3 Nov 2026: “Kling 4.0, the new-generation video generation model of Kling AI, officially launched on October 15.” · Kling AI API documentation, Kling 4.0 Video User Manual, banner read in a browser 3 Nov 2026: “[New Release] Kling 4.0 API Is Now Officially Live and Ready for Integration!” · Kling AI, Inside Kling 4.0: A Complete Guide to AI Video Creation, 30 Sep 2026, re-read at source 3 Nov 2026, not updated since: “The full Kling 4.0 model is scheduled to launch in October 2026.” ↑
- 24Runway, Creating with Gen-4.5, read at source 22 Sep 2026“Cost 12 credits per second”. ↑
- 25Black Forest Labs, FLUX.2 launch post, 25 Nov 2025, read at source 16 Sep 2026. bfl.ai↑
- 26Black Forest Labs, Release notes, 4 Aug 2026 entry, read at source 22 Sep 2026“FLUX 3 is our first video model”, “Available now as a preview.” · API Pricing, read at source 22 Sep 2026: “Pay per image, or per second for video”. ↑
- 27Black Forest Labs, Release notes, 10 Sep 2026 entry, read at source 1 Oct 2026“a 2K or 4K clip comes back from a single request instead of a generation followed by a separate upscale call.” ↑
- 28Midjourney, Version, read in a browser 17 Sep 2026. An earlier version of this card presented “Midjourney v7 (April 2026)” as current. Midjourney: “V7 was released on April 3, 2025, and was the default version from June 17, 2025 to June 9, 2026.”↑
- 29Udio Help Center, Changes associated with the Universal Music Group partnership, read at source 22 Sep 2026“Note that downloading of audio, video, and stems has been disabled”. ↑
- 30Udio, Introducing v1.5, read at source 22 Sep 2026.↑
- 31Suno, v6 release notes, 9 Sep 2026, and v6 FAQ; read at source 16 Sep 2026.↑
- 32Suno, How to make a song and How do I exclude elements of a song?, read at source 16 Sep 2026. An earlier version of this card said the style box wants comma-separated descriptors, not sentences, and gave third-party character limits and a descriptor sweet spot for v4.5–v5.5, models Suno has since retired. Those were removed.↑
- 33ByteDance Seed, Introducing Seedream 5.0 Pro, 8 Jul 2026, read at source 16 Sep 2026“point selection, lasso selection, sketch rendering, color and material replacement, layer separation, and multi-image fusion”. An earlier version of this card said editing works "rather than layers" and dated the release 10 February 2026; Dreamina's Seedream 5.0 guide is dated 5 February and ByteDance's 5.0 Lite post 13 February. ↑
- 34Dreamina (ByteDance) Seedream 5.0 Pro guides, dreamina.capcut.com, checked 25 Aug 2026↑
- 35ElevenLabs, Eleven v4, 28 Sep 2026, read at source 1 Oct 2026speech that can sound “dramatic, tender, urgent, comedic, or conversational”; “More than 90 languages”; available in “ElevenAgents, ElevenCreative, and via ElevenAPI”. ↑
- 36ElevenLabs prompting documentation, elevenlabs.io/docs, checked 25 Aug 2026↑
- 37Higgsfield, Changelog, read at source 1 Oct 2026, 1 Sep 2026 entryGenjutsu “changes an object, location, product, or character in a video while leaving everything else as it was.” ↑
- 38Pollo AI, Plans & Pricing, read in a browser 23 Sep 2026compare table, “Free”, “20 credits”; Pollo 2.5 720P, “5 credits/5s”, “4 videos”. The rendered page does not mention daily credits. ↑
- 39Build Fast with AI, Google Veo 3.1 Review (2026), read at source 23 Sep 2026“Veo 3.1 is the first practical AI video model to generate synchronized audio at 48kHz directly from a text prompt, with lip-sync accuracy within 120ms.” Pinggy, Best Video Generation AI Models in 2026, updated 14 Jul 2026: “Veo is the only widely available model generating 48kHz synchronized dialogue”. Both are third-party reviews, not Google documentation; first checked 22 Aug 2026 against these and oTechWorld. Until 23 Sep 2026 this line said “verified”. ↑
- 40Google, Generate videos with Veo 3.1, read at source 1 Oct 2026lengths “8 seconds, 6 seconds, 4 seconds”; “up to 148 seconds of video”; “limited to 720p videos”; “Videos created by Veo are watermarked using SynthID”. ↑
- 41Google, Gemini Omni (Gemini app overview), undated, read at source 13 Oct 2026“Gemini Omni is our latest video editing and generation model that will replace Veo in the Gemini app.” ↑
- 42Google, Flow updates, 19 May 2026, read at source 29 Sep 2026“blend real-world inspiration with generated content and iterate conversationally” · Google, Introducing Gemini Omni, 19 May 2026, read at source 16 Sep 2026: rolling out “through the Gemini app and Google Flow”. ↑
- 43Google, Gemini API deprecations, last updated 9 Oct 2026, read at source 13 Oct 2026the three Imagen 4 models, shutdown “August 17, 2026”, replacement “gemini-nano-banana-2.1”;
gemini-3.1-flash-image, “No shutdown date announced”. Read at source 16 Sep 2026, this page and the models page namedgemini-3.1-flash-image(Nano Banana 2) as the replacement, and until 13 Oct 2026 this card did too. An earlier version of this card listed Imagen 4 as current, checked 22 Aug 2026 — five days after its shutdown date. ↑ - 44Google, Gemini API release notes, 6 Oct 2026, read at source 13 Oct 2026“Released Gemini Nano Banana 2.1 (gemini-nano-banana-2.1), the latest high-efficiency image generation and conversational editing model.” ↑
- 45Google, "NotebookLM is now Gemini Notebook", announced 16 July 2026, read at source 11 Sep 2026“Gemini Notebook and NotebookLM are the same product. You do not need to move your notebooks, recreate your sources, or learn a replacement tool”. Previously cited here to Man of Many, Aug 2026. ↑
- 46Man of Many, Aug 2026 · GitHub list-of-free-google-ai-tools · Wikipedia, Google Antigravity. Whisk and ImageFX: Google, Flow updates, 25 Feb 2026, read at source 16 Sep 2026; labs.google/whisk and labs.google/fx/tools/image-fx both redirect to flow.google.com, checked 16 Sep 2026. An earlier version said both “closed in April 2026”, a date no Google source gives.↑
- 47OpenAI Help Center, What to know about the Sora discontinuation, read in a browser 17 Sep 2026“The Sora web and app experiences were discontinued on April 26, 2026.” and “The Sora API will be discontinued on September 24, 2026.” ↑
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK