FLUX 3
FLUX 3 is a video model family by Black Forest Labs. FLUX 3 in Vini generates from text or a first frame and edits clips. It has no separate hero-reference mode in this connection. Put the intended character and visual style into the approved first frame when they must anchor the result.
Resolution, quality and sound
Section titled “Resolution, quality and sound”FLUX 3 has three variants with the same inputs: Draft (черновик), Full (полное) and Max. Only the address, the price and the resolution list change. Draft generates at a fixed 720p. Full offers 720p, 1080p, 1440p (2K) or 2160p (4K); the provider finishes 1080p and higher with its video upsampler, so those options should not be read as native generation resolutions. Max is the provider’s post-trained FLUX 3 variant and offers 720p or 1080p; it covers text-to-video, first-frame generation (a last frame is accepted) and keyframes. Editing has no variant choice.
Since 25 October 2026 new nodes use Max at 720p, with sound on, because it is currently the cheapest variant; the card shows the cost before the run. Max asks the provider for its lowest prompt-reasoning level, so the provider rewrites the prompt as little as its controls allow.
Generation accepts 5–20 seconds and supports turning sound off. Editing follows the source and has no separate resolution, quality or sound control. Saved explicit choices remain selected; an older node without a quality value retains its Full fallback.
Settings were checked against the FLUX 3 API and Draft API on 20 October 2026, and the FLUX 3 Max API on 25 October 2026. Higher output resolution does not establish a new visual-quality or identity-retention result, and the Max variant has no studio quality exam yet.
| Setting | Value |
|---|---|
| Model ID | flux3 |
| Name | FLUX 3 |
| Availability | available |
| Default for new nodes | no |
| Earlier IDs resolving to this record | flux3_draft |
Common declared limits
Section titled “Common declared limits”| Setting | Value |
|---|---|
| Capabilities | text to video, image to video, native audio, dialogue |
| Prompt required | yes |
| Maximum prompt length | no fixed maximum declared |
| Duration | 5–20 seconds; fallback 5 seconds |
| Audio | model-generated audio, off; fallback model-generated audio |
| Aspect ratio | 9:16, 16:9, 1:1, 21:9, 2:1, 4:3, 3:4; fallback 9:16 |
| Declared language support | prompt: English; speech: English, Russian; tier: platinum |
| Controls | resolution: 720p, 1080p; fallback 720p; quality: draft, full, max; new node max; fallback full |
| Element input form | sheet |
| Content constraints | people: restriction level: ok |
| Negative prompt strategy | instruction in prompt text |
Mode: from_first_frame
Section titled “Mode: from_first_frame”| Setting | Value |
|---|---|
| Keyframes | maximum: 10; positions: free; unit: seconds |
| First frame | required |
| Last frame | accepted; optional |
Quality: draft in from_first_frame
Section titled “Quality: draft in from_first_frame”| Setting | Value |
|---|---|
| Controls | resolution: not declared; quality: draft, full, max; new node max; fallback full |
Quality: full in from_first_frame
Section titled “Quality: full in from_first_frame”| Setting | Value |
|---|---|
| Controls | resolution: 720p, 1080p, 1440p, 2160p; fallback 720p; quality: draft, full, max; new node max; fallback full |
Mode: text_to_video
Section titled “Mode: text_to_video”Uses the common declared limits above.
Quality: draft in text_to_video
Section titled “Quality: draft in text_to_video”| Setting | Value |
|---|---|
| Controls | resolution: not declared; quality: draft, full, max; new node max; fallback full |
Quality: full in text_to_video
Section titled “Quality: full in text_to_video”| Setting | Value |
|---|---|
| Controls | resolution: 720p, 1080p, 1440p, 2160p; fallback 720p; quality: draft, full, max; new node max; fallback full |
Mode: edit_video
Section titled “Mode: edit_video”| Setting | Value |
|---|---|
| Maximum prompt length | 4096 |
| Duration | same as source video; fallback same as source video |
| Audio | not declared |
| Aspect ratio | not declared |
| Controls | resolution: not declared; quality: not declared |
| Source video | required |
Unavailable modes: avatar_audio, extend_video, from_references, motion_transfer, recast.
Limits apply together. initial is the studio choice for a new node; default is the fallback for an existing node without that value. Declared language or input support does not guarantee a successful result.
Prompting playbook
Section titled “Prompting playbook”- Describe what happens over time, including the interaction between objects, rather than an inventory of objects.
- Use an approved scene image as the start, not a multi-view identity sheet.
- For edits, name the changed object and the scene properties that must remain. A fresh Full generation is not a promise to reproduce a Draft result.
The built-in FLUX 3 authoring skill adapts a scene request for this family through Prompt for the model. It prepares instructions; it does not render video or certify the result.
Studio example
Section titled “Studio example”Prepare a first frame in the desired illustrated style, showing a toy boat beside a puddle. Select FLUX 3 first-frame mode. Ask Prompt for the model: “The toy boat drifts across the puddle as a leaf lands beside it. The camera follows slowly at water level.”
A run may consume credits. Review the prompt and inputs, then inspect the finished clip for action, identity, framing and sound. For automation, use the supported MCP workflow.
Known failure modes
Section titled “Known failure modes”Without a connected hero-reference mode, adding more character descriptions does not create an identity-binding input. The model author’s continuation and draft-enhancement features do not imply corresponding controls in Vini. Motion requested from an attractive first frame can still be incomplete.
Evidence and limits
Section titled “Evidence and limits”On 4 October 2026, a Draft first-frame test kept the prepared stylized appearance throughout the clip, but did not complete the entire requested walk. No dedicated current character-reference exam exists for this model because that mode is not connected. Text-only video editing has not received the same formal studio exam as the tested reference-capable models.
The studio registry establishes available inputs and modes. A recorded render establishes only its observed outcome. The applicability guide explains the distinction. See also character identity, static camera and video style samples.
See the model–technique applicability matrix for connected operations, dated outcomes and configurations that remain untested.
Reviewed 14 October 2026 against the current studio contract and dated studio research. Official model-author documentation checked on that date; author capabilities may exceed Vini’s connected controls. Recheck this playbook when the model version, quality or mode changes.