Sources · checked October 20, 2026
How to write FLUX 3 Max prompts
Write a FLUX 3 Max prompt in five parts: one camera move, the main action as beats in order, secondary motion in the environment, a change in light, and a closing Sound line. fal derived this shape from 1,000 test clips; for image-to-video, spend the words on motion, because the start image already sets the look.
The five parts
| Part | What to write | fal's example |
|---|---|---|
| Camera | One move and its speed; give a second move its own verb | "Static symmetrical frame." |
| Main action | Beats in order, linked with "then" or "as" | "The fjord lies glassy and still; a faint breath of wind sends a slow ripple across the water…" |
| Secondary motion | Something in the environment that moves on its own | "The tied rowing boat rocks almost imperceptibly on its line." |
| Light | What changes in the light, and what stays still ("Nothing else moves") | "Soft clouds drift across the pink sky." |
| Sound | Three or four specific sounds | "Sound: near silence, a faint lap of water, a distant bird call." |
Adjust the prompt to the endpoint
- Text-to-video: describe a clear sequence of visible actions. Camera, lighting and atmosphere define the shot; dialogue or ambient sound defines the audio.
- Image-to-video: describe what changes after the opening frame. Point at what is already in the picture with "the" ("the crow on the rim") instead of describing it again.
- Keyframes-to-video: describe the movement between the pinned frames. Keyframes guide the sequence but do not fix every detail between them.
Dialogue and sound
- Put the exact words in quotation marks and name who says them. A visible speaker gets lip-sync; an off-screen voice needs "voiceover" or "narration".
- Name specific sources ("rain ticking against a metal awning") instead of mood words, and name a quiet sound instead of asking for silence.
- If a spoken line appears as on-screen text, add "no on-screen text or subtitles". Set generate_audio to false if you will add your own sound.
These audio rules come from Black Forest Labs' FLUX 3 Video guide, the model FLUX 3 Max is post-trained from.
Example prompts published by fal
A red panda walks along a mossy log in a sunlit forest, pausing to sniff the air before climbing over a fallen branch. Soft morning light filters through the canopy. Ambient birdsong and rustling leaves.
The monkey leans into his microphone and says, "Welcome back to the show!" The polar bear laughs and nods along, tapping the desk. Jungle birdsong on the left, a cold wind on the right.
Static symmetrical frame. The fjord lies glassy and still; a faint breath of wind sends a slow ripple across the water, gently breaking the mirrored red boathouse and the green hills before they settle back into perfect reflection. The tied rowing boat rocks almost imperceptibly on its line. Soft clouds drift across the pink sky. Sound: near silence, a faint lap of water, a distant bird call.
Questions and answers
Should I describe everything in my start image?
No. With image-to-video the start image already sets the look, so fal's guide spends almost every word on motion and refers to things in the picture with "the".
How do I stop FLUX 3 Max from cutting to a new angle?
Ask for one continuous shot in the prompt. fal saw the model cut to a new angle partway through in a few of the 1,000 takes it reviewed.
What does reasoning_level do?
It sets how much effort FLUX 3 Max spends reasoning about your prompt: high, the default, is slower; low is faster. fal does not return a rewritten prompt.
Is there a prompt length limit for FLUX 3 Max?
No limit is documented by fal or Black Forest Labs. BFL warns that over-stuffing a prompt can make motion less coherent, so start short and add only the details that matter.