ONLINEAGENT_OPS 2026.Q3 HOME ARTICLES CRAFT RECORD BLOG MAP HUBS FAQ SEARCH
HOMETHE CRAFTOFFICIAL FORMULAS
THE CRAFT · VENDOR DOCUMENTATION

What the Vendors Actually Say

The prompt formula each vendor publishes for its own model, from FLUX and Veo to Kling, Midjourney, Suno and Higgsfield, and which ones take a negative prompt.

READ11 min
WORDS2,257
SECTIONS4
SOURCES11
TYPEGUIDE
CHECKED16 SEP 26

Ten tools, and each vendor publishes its own idea of what a prompt is. They do not agree — and the disagreements are the useful part.

TL;DR — THE SHORT VERSION
  • Four of seven models take no negative prompt at all. FLUX.2, Runway, Seedance and ChatGPT Image have no field. Runway documents that using one may produce the opposite.
  • No two formulas open the same way. The subject comes first for FLUX.2 and Kling; Google opens with the camera, ByteDance’s Seedance with format, OpenAI with the scene, and Runway with the one essential motion.
  • Word order carries weight. Black Forest Labs states plainly that FLUX.2 attends most to what comes first.
  • 30–80 words is the documented sweet spot for FLUX.2. Longer is for genuinely complex scenes, not for more obedience.
  • In a negative field, keep entries short and specific. Google’s own Veo example keeps the word “no”, and Kling recommends negative sentences inside the prompt.
  • OpenAI rewrites your prompt before generating it. What you typed is not what the model received.
  • Audio has no prompt box. On ElevenLabs the script is the prompt — punctuation and capitalisation are the controls, and v3 and v2 take incompatible tag syntax.
  • Suno splits the prompt. Style and lyrics are separate fields, and what you do not want goes in Exclude.
  • ByteDance publishes an eight-part Seedance formula that puts format first and action third — not the motion-first bracket format most guides teach.

The table nobody prints

Every negative-prompt guide online hands you one list and implies it works everywhere. Check the vendor documentation and that collapses immediately. This is the single highest-value thing on this page.

FLUX.2 — "does not support negative prompts"NO FIELD
Runway Gen-4 / 4.5 — unsupported; "may result in the opposite happening"BACKFIRES
Seedance 2.5 — no field; exclusions compete with your positive promptNO FIELD
ChatGPT Images — no field; exclusions go inside the promptNO FIELD
Midjourney v8.2--no at the end, items separated by commasYES
Kling 3.0 — separate field on the legacy API only; Kling recommends negatives inside the promptLEGACY
Veo 3.1negativePrompt on Gemini Enterprise Agent Platform, formerly Vertex AIYES
Vendor documentation: docs.bfl.ai · help.runwayml.com · Luma and Melies Seedance guides · developers.openai.com · Google Cloud Veo prompting guide · Kling API documentation. Checked 25 Aug 2026. Veo parameter re-read 16 Sep 2026 in Google Cloud’s documentation, where Vertex AI now appears as Gemini Enterprise Agent Platform.
TAKEAWAY

Before you write a negative prompt, find your model in that table. On four of seven, the correct move is not a shorter negative — it is no negative at all, with every exclusion rewritten as a description of what you want present.

The formulas, as published

Image and video first, then text, then audio — one cheat sheet per model. Where a vendor publishes no consolidated guide — Midjourney, Suno, Higgsfield — the entry says so and the sourcing is flagged as weaker.

WHAT COMES FIRST · SIX PUBLISHED FORMULAS
The first three slots of each formula as printed below, subject highlighted. It is first for FLUX.2 and Kling, second for Veo, Seedance and ChatGPT Images; Runway starts from the essential motion and adds elements one at a time.
FIRSTSECONDTHIRDFLUX.2SubjectActionStyleVeo 3.1CinematographySubjectActionRunwayEssential motionSubject motionCamera motionSeedance 2.5FormatSubjectActionKling 3.0SubjectSubject MovementSceneChatGPT ImagesSceneSubjectDetails
Reasoning — lays out the formulas as printed in this page’s sections for FLUX.2, Veo 3.1, Runway, Seedance 2.5, Kling 3.0 and ChatGPT Images, each sourced in its own section, page checked 17 Sep 2026.
01

FLUX.2 — Black Forest Labs

Subject + Action + Style + Context

The distinctive claim is about order, not content: the model attends most to what comes first. Their stated priority is main subject → key action → critical style → essential context → secondary details. Bury the subject at the end of a long sentence and you are working against the model.

Length: 10–30 words for quick concepts; 30–80 for most work; 80+ only for complex scenes. They also support a JSON prompt schema for production use, and note that hex colour codes only work reliably when attached to a specific object — a stray code with no object attached will not hold.

Black Forest Labs, FLUX.2 [pro] & [max] prompting guide, re-read 11 Sep 2026: “FLUX.2 does not support negative prompts”; “instead, focus on describing what you want, not what you don’t want”
02

Veo 3.1 — Google

[Cinematography] + [Subject] + [Action] + [Context] + [Style & Ambiance]

The only major formula that opens with the camera rather than the subject. Shot size and camera move come first, then who, then what they do.

Audio has its own syntax, which most guides miss entirely — dialogue goes in quotation marks, effects are prefixed SFX:, and background is prefixed Ambient noise:.

On negatives: Veo has the field, but Google's instruction is to describe the exclusion as something present. Their own example — write "a desolate landscape with no buildings or roads", not "no man-made structures".

Google Cloud, Ultimate prompting guide for Veo 3.1, re-read 11 Sep 2026: “to refine your output, describe what you wish to exclude”
03

Runway Gen-4 / 4.5

Essential motion first → then one element at a time: subject motion, camera motion, scene motion, style

Start simple. Runway’s video guide: “Begin with a foundational prompt that captures only the most essential motion to the scene.” For images it says “Simple prompting works well, but more complex prompt will yield more stylistic control”, and recommends “Use full sentences with natural language for more control over elements”.

The hard rule: "Avoid negative prompting, such as no clouds in the sky, for the best prompt adherence." And the reason it matters — their documentation states that including a negative may result in the opposite happening. This is the one model where the folk warning is the vendor's own.

Runway, Gen-4 Image Prompting Guide and Gen-4 Video Prompting Guide, read in a browser 17 Sep 2026: negative prompts “are not supported in Gen-4 Images. Including a negative prompt may result in the opposite happening”. An earlier version of this entry gave Runway’s formula as Subject → Action → Setting → Camera → Motion over time → Style; the guides do not give that order.
04

Seedance 2.5 — ByteDance

Format + Subject + Action + Environment + Camera + Look + Timing + Audio & Constraints

ByteDance publishes this eight-part structure through Dreamina, its own platform. It is not the bracket format most guides teach — including, until now, this site. Format comes first (duration, aspect ratio, shot structure), the subject second, and the action third, not first.

Their published template: [Duration, aspect ratio, shot structure]. Use [@reference] as the consistent [character/product/location]. In [environment and time], the subject [precise action]. Camera: [shot size + movement + lens/focus]. Visual direction: [lighting + palette + texture + realism]. Timing: [beat 1] → [beat 2] → [beat 3] → [resolution]. Audio: [dialogue/ambience/SFX/music]. Preserve [identity, wardrobe, logo, object geometry]. Avoid [unwanted cuts, camera behavior, artifacts, extra objects, text].

Four rules inside that template that most guides miss:

Bind references with @Image1, @Image2. Tag assets to specific beats rather than hoping the model infers which reference governs what.
Shot size before style language, and one camera move per shot. Two moves in one shot is a contradiction the model has to resolve.
Roughly 6–8 seconds per beat is the stated default for timing.
Preserve, then avoid. The template gives exclusions their own slot, short, concrete and grouped at the end — after a positive Preserve [...] clause. That ordering matters: state what must survive before what must not happen.

On negatives: there is no negative_prompt field, so exclusions share the prompt and compete with it. Long generic lists contradict the positive direction — banning blur while asking for shallow depth of field, banning shake while asking for handheld. Keep the Avoid clause to a handful of unusable-clip failures.

Dreamina (ByteDance), Seedance 2.5 prompt guide, dreamina.capcut.com — the vendor's own platform, checked 25 Aug 2026. Supplementary: Luma Labs and Melies Seedance guides, which agree on short targeted exclusions. Correction to an earlier version of this page, which described Seedance sourcing as third-party and named the bracket format as the vendor structure. ByteDance publishes a formula; this is it.
05

Kling 3.0 — Kuaishou

Subject + Subject Movement + Scene — and negatives inside the prompt

Kling's own formula is Subject (with a description) + Subject Movement + Scene (with a description), with camera language, lighting and atmosphere as optional extras. A separate negative_prompt field survives only on Kling's legacy API — and even there Kling says “It is recommended to supplement negative prompt via negative sentences within positive prompts”. Its upgraded API has no separate field: “The prompt can include positive and negative descriptions”.

Kling AI, Text-to-video prompt guide and Kling AI API, Text to Video (legacy), read 16 Sep 2026, the API reference in a browser. An earlier version of this entry called it "a real negative field" where Stable Diffusion habits "transfer cleanly"; Kling's own documentation points the other way. Full Kling sheet.
06

Midjourney v8.2

--no at the end, items separated by commas

Midjourney: “Add --no to the end of your prompt in the Imagine bar, followed by the thing or list of things you don't want in your image.” Several items go in one list: “You can even list multiple elements by separating them with commas.”

It is a weight, not a ban. “Using the --no parameter is the same as weighing part of a multi-prompt to "-0.5"” — so treat it as a strong push away from something, not a guarantee it will not appear.

Midjourney, No parameter, read in a browser 16 Sep 2026. An earlier version of this entry said Midjourney's docs could not be retrieved, that a second --no does not stack, and that each word after --no is read separately. The docs are public; the other two claims are not in them and were removed. Full Midjourney sheet.
07

ChatGPT Images — OpenAI

Scene, subject, details, constraints — and your prompt may be rewritten

OpenAI's image prompting guide gives a structure: “For complex requests, organize the prompt as scene, subject, details, and constraints, using labeled sections.” Exclusions go inside the prompt: “State exclusions such as unwanted text, logos, or watermarks.”

On the Responses API the main model “will automatically revise your prompt for improved performance”, and the rewrite comes back in revised_prompt. When output does not match your words, read that first. OpenAI also says the model “can still struggle with precise text placement and clarity”.

OpenAI, Image prompting guide and Image generation guide, read at source 16 Sep 2026. An earlier version of this entry said OpenAI publishes no prompt formula. Full ChatGPT Images sheet.
08

ElevenLabs — voice

The text is the prompt. There is no separate prompt box.

Punctuation and capitalisation do real work. ElevenLabs: “Ellipses (…) add pauses and weight”, “Capitalization increases emphasis”, and “Standard punctuation provides natural speech rhythm”.

The model split matters more than the tags. Eleven v3 takes inline audio tags — [whispers], [sighs], [sarcastic] — and “Eleven v3 does not support SSML break tags.” Models that do support them take <break time="1.5s" /> for pauses up to 3 seconds, and ElevenLabs warns: “Using too many break tags in a single generation can cause instability.”

The limit that catches people: a tag cannot override the voice. “Don’t expect a whispering voice to suddenly shout with a [shout] tag.” Professional Voice Clones are “currently not fully optimized for Eleven v3”.

ElevenLabs, Text to Speech best practices, read at source 16 Sep 2026. An earlier version of this entry said syntax from the wrong model "fails silently"; ElevenLabs does not say that. Full ElevenLabs sheet.
09

Suno — music

Style, lyrics with structure tags, and Exclude — on v6

The style prompt carries the sound. Suno: “To write a prompt, be specific: mention genre, mood, keywords, and instrumentation.” Its own example is written in sentences: “Style: Contemporary R&B song at 92 BPM in F minor.” The lyrics field carries the words, and “Suno works best when you guide it with structure tags such as [Verse], [Chorus], and [Bridge].”

What you do not want goes in Exclude, in Custom Mode under Advanced Options. Every model before v6 is retired: “All models prior to v6 have been retired, but your songs will still be in your library and remain unchanged.”

Suno: How to make a song, How do I exclude elements of a song?, v6 FAQ; read at source 16 Sep 2026. An earlier version of this entry said the style box wants comma-separated descriptors rather than sentences, and gave third-party character limits and a descriptor "sweet spot" for models Suno has since retired. Those were removed. Full Suno sheet.
10

Higgsfield — motion

Name the move in the prompt and select the matching preset.

Higgsfield's distinguishing feature is a library of 50+ named cinematic motion presets — crash zoom, bullet time, dolly, crane, FPV drone, Snorricam. The workflow is not prompt-only: you pick the preset in the app, then describe subject and scene in the prompt. Naming the same move in the text as well as selecting it is what makes results repeatable rather than approximate.

Because the camera move comes from the preset, the prompt's job is narrower than on other video tools: subject, identity anchoring, environment and action. Describing an elaborate camera move in text while a different preset is selected puts the two in conflict — the same self-contradiction problem that breaks long negative prompts elsewhere.

Higgsfield camera-controls preset library, higgsfield.ai/camera-controls, checked 25 Aug 2026. Higgsfield does not publish a single consolidated prompting guide; the workflow above is drawn from the preset documentation and platform guides.

Where the vendors agree, and where they split

Strip the branding and three rules are left. The label on each says how far the vendors actually agree:

Say exactly what to leave out. How varies: Black Forest Labs and Runway want every exclusion rewritten as something present; Google’s Veo example keeps “no” and names the absence; Kling says “The prompt can include positive and negative descriptions”.SPLIT
Put what matters early. Explicit for FLUX.2: “Word order matters - FLUX.2 pays more attention to what comes first.” The other ordered formulas each fix an opening slot (camera, format, scene or subject), and Runway starts from the essential motion.MOST
Specific beats long. Nobody recommends maximum length. Several warn against overload — ElevenLabs says too many break tags “can cause instability”, and ByteDance warns against “stuffing too many styles, ideas, or conflicting instructions into one prompt”.ALL

The checklist

Before you send a prompt:

1 — Found my model in the negative-prompt table above?
2 — Subject early, in the slot your vendor’s formula gives it?
3 — Between 30 and 80 words, unless the scene genuinely needs more?
4 — Exclusions handled the way your vendor says — rewritten as something present, or kept short inside the prompt?
5 — If a negative field exists: short, specific entries?
6 — Nothing in the negative contradicting the positive?
7 — Video: camera move, subject and duration each stated once?
8 — Veo: dialogue in quotes, SFX: and Ambient noise: prefixed?
HOW TO CHECK THIS PAGE YOURSELF

Every claim here traces to a vendor's own documentation, named in the source line under each section. Model versions and documentation both move — this page was checked on 25 August 2026, and its Kling, Midjourney, ChatGPT Images, ElevenLabs and Suno entries were re-checked at source on 16 September 2026 and the negative-prompt table is the part most likely to change first. Re-read the vendor page before relying on any row of it. Errors are corrected in place on this page, with a note where the change matters.

Where vendors contradict each other, this page says so rather than picking the tidier answer — see the Seedance note above.

The through-line: prompting is not one skill applied to ten tools. The tools disagree about what a prompt even is, and the vendors have written that down.

ABOUTMETHODVERIFYPRIVACYCONTACTINDEXAI PROMPT GENEER · EVERY ARTICLE CARRIES ITS OWN CHECKED DATE