Black Forest Labs announced the multimodal FLUX 3 family on July 23, 2026. Twelve days later, on August 4, FLUX 3 Video reached general availability. Those two dates are useful not only as release milestones, but as evidence of a wider change in how visual tools are described: image making, editing, and motion increasingly belong to one connected production conversation.
For image creators, “multimodal” should not be reduced to a feature label. It changes how a brief is prepared, how frames are evaluated, and how visual continuity is preserved when an idea moves between a key image, a campaign crop, and a timed sequence. The valuable response is to build a workflow that survives the transition between formats.
Two dates with different meanings
An announcement date establishes the family, its direction, and the language the publisher uses to describe it. A general-availability date establishes a product surface that teams can put into a production plan. Keeping those milestones separate makes technical reporting clearer and helps creative teams avoid treating every demonstration as an immediately repeatable workflow.
The July 23 announcement matters because Black Forest Labs framed FLUX 3 as multimodal. That points toward a visual system in which the meaning of a prompt, reference, and sequence can matter across more than one output type. The August 4 Video release matters because it converts part of that direction into a priced API product with concrete operational choices.
For creators, the practical lesson is simple: attach every claim to a product and a date. A family-level idea can inspire a workflow, while a product-level release defines the inputs, prices, limits, and outputs a team can actually test. Both are useful, but they answer different questions.
Multimodal work begins with a stronger brief
A production brief should describe the visual invariants that must survive across formats. Start with the subject, environment, physical relationships, materials, light, and camera position. Then identify what can change: crop, duration, motion, expression, time of day, or the order in which details are revealed.
Imagine a campaign built around a compact field recorder on a dark cork desk. The stable decisions might be the recorder’s three-quarter angle, the direction of soft window light, the anodized aluminum texture, and a quiet charcoal palette. A square product image, a wide page header, and a short motion study can all vary while preserving those decisions.
This is more durable than a prompt made from mood adjectives. “Premium, cinematic, beautiful, highly detailed” gives reviewers little to inspect. “Three-quarter front view, soft window light from camera left, realistic anodized aluminum, restrained reflections, clear space on the right” creates observable criteria that can be carried into another frame or medium.
Treat a still image as a keyframe
Image creators already make decisions that resemble keyframe direction. A still establishes subject scale, lens language, depth, light, palette, and the relationship between foreground and background. When motion becomes part of the job, those decisions become the visual anchor of a sequence.
That does not mean every image should look like a film still. A product tile may need flat clarity and a centered silhouette. An editorial illustration may rely on graphic simplification. The point is to understand which properties define the visual system, so movement does not accidentally replace the art direction.
Write down the camera position and what is allowed to move. Note whether light should remain fixed or change with time. Identify elements that must not enter the frame. If text or a product label matters, decide whether it belongs in generated imagery or should be applied later in a controlled design tool. These decisions make transitions intentional.
Aspect ratio is part of the idea
A square, 4:3 frame, 16:9 banner, and 9:16 mobile composition are not interchangeable containers. Each one changes the space available around a subject and the path a viewer’s eye takes through the image. Cropping after the fact often removes precisely the context that made a composition work.
Choose the destination before generation. A wide campaign image may need negative space for a headline. A vertical mobile frame may need the subject below interface controls. An editorial opener may need a focal point that survives responsive crops. A storyboard frame needs continuity with the shots around it.
When one concept must serve several formats, preserve the visual invariants and compose each important orientation deliberately. This produces a family of related assets rather than several damaged crops of one master image.
Evaluate continuity with observable criteria
Visual continuity is easier to discuss when the review language is concrete. Compare camera height, horizon, subject scale, shadow direction, material response, color temperature, depth of field, and the placement of negative space. These properties can be inspected across still images and timed sequences.
Avoid review notes such as “make it more epic” or “it needs more energy.” Translate them into visible changes. More energy might mean a lower camera, a stronger diagonal, closer subject scale, harder side light, or motion that crosses the frame. Each interpretation creates a different result, so the reviewer should choose.
Save accepted prompts and rejected outputs together. Record the largest problem in each rejected version. This creates a compact evaluation set for future work and prevents a team from repeating the same failed direction after the original discussion has been forgotten.
Price video by the sequence, not the clip
FLUX 3 Video’s published full-render prices are $0.17 per second for HD and $0.29 per second for FHD. Draft generation is $0.06 per second, and HD video continuation is $0.43 per second. Those units reward planning because duration and iteration depth directly affect cost.
A creator should therefore budget the exploration path, not only the final clip. Use still frames and storyboards to settle composition before paying for motion. Use draft output to test timing or camera direction where its quality is sufficient for a decision. Reserve final-resolution runs for sequences whose structure already works.
Continuation also deserves its own budget. Extending a clip is not simply “more of the same” when it is priced at a different rate. Decide what the extension must accomplish, how many seconds it needs, and which visual invariants must remain intact.
Keep rights and review in the workflow
Multimodal capability does not create permission. Reference images, recognizable people, trademarks, artworks, confidential product designs, and client material still require appropriate rights and handling. A visual transformation can introduce new risks when it makes a person appear to act, speak, or endorse something they did not.
Review generated work in its final context. Check labels, logos, anatomy, repeated objects, reflections, architecture, culturally sensitive details, and factual implications. For motion, inspect temporal continuity as well: objects should not change identity, disappear without reason, or acquire misleading behavior between frames.
Important outputs should be downloaded and preserved with their brief, aspect ratio, reference version, and intended use. A hosted history is convenient, but it should not be the only record of a production asset.
The durable advantage is a portable visual system
The strongest response to a new visual model family is not to chase a secret phrase. It is to make the creative input more portable. A clear brief, explicit invariants, destination-aware framing, saved references, and observable review criteria remain useful when the output format changes.
The July 23 announcement and August 4 Video release make the direction visible: visual production is becoming more connected across still and motion work. Creators who treat prompts as specifications and images as parts of a larger system will be better prepared than those who treat every generation as an isolated surprise.