FLUX 3 on Promptus

Black Forest Labs released FLUX 3 on July 23, 2026 — its first model trained jointly across image, video, and audio in a single architecture, rather than as separate systems bolted together. This article covers what FLUX 3 actually does, how it compares to other video models based on Black Forest Labs' own early testing, and everything currently known about accessing it through Promptus, including pricing and API details.

What FLUX 3 actually is

FLUX 3 is a multimodal foundation model that learns from images, video, and audio at the same time, within one architecture, rather than training separate models for each. Black Forest Labs' reasoning: sound has to match physical impact, motion has to obey mass, and a model that only ever sees still images never learns any of that. Training on all three together, in their framing, means each modality's constraints teach the other two — the model isn't guessing at motion and sound separately, it's learning one internal representation that produces all of it together.

Practically, this shows up as native audio generation — FLUX 3 Video generates synchronized sound in the same pass as the video, not as a separate step stitched on afterward.

What FLUX 3 can do

Confirmed capabilities, per Black Forest Labs' own announcement:

• Text-to-video with native audio, generating clips up to 20 seconds long with synchronized dialogue, sound effects, and ambient sound.
• Image-to-video, in two distinct modes: animating from a fixed starting frame, or using a reference image to guide the scene without pinning the exact first frame.
• Video-to-video and continuation, carrying central elements — including a specific character — from an existing clip into a new scene.
• Keyframe-to-video, generating a controlled transition between two defined moments rather than an open-ended animation.
• Multilingual dialogue and facial expression, generated directly rather than dubbed afterward.
• Broad style and aspect-ratio range, from candid, camcorder-style footage to full cinematic composition, plus strong on-screen typography and animated text.
• Agentic multi-shot chaining, linking individual generated clips into longer sequences — though this is a sequence of clips, not a single continuous generation; the native clip length remains up to 20 seconds per generation.

Image generation and action-prediction capabilities (the latter aimed at robotics, in partnership with mimic robotics) are part of the same announcement but are on a separate rollout timeline — image access is expected in the following weeks, not immediately.

How FLUX 3 compares to other video models

Black Forest Labs published preliminary preference testing, comparing an early FLUX 3 candidate against current video models. In their own evaluation, using 10-second, 720p text-to-video clips:

Important context on these numbers: this is Black Forest Labs' own internal testing, not an independent benchmark. The published results don't include the full prompt set, evaluator instructions, sample count, or confidence intervals, and the test used shorter, lower-resolution clips than FLUX 3's full advertised range. Treat this as a strong early signal from the model's own developer, not a verified third-party ranking. We'll update this section if and when independent testing becomes available.

FLUX 3 on Promptus — current status

Pending. As of this writing, FLUX 3 Video is not yet listed as an available model in Promptus. Black Forest Labs is rolling out FLUX 3 Video through API and private-weight access first, with broader availability to follow — this is not yet an instant, self-serve public endpoint on their end either. Promptus will add FLUX 3 to the model picker on its own schedule once integration is complete, alongside its own pricing, supported modes, and any generation limits.

Once live, FLUX 3 will appear in Promptus's Playground under the Video task, in the model selection menu. We'll confirm the exact process here as soon as it's available — including whether it launches as a cloud-credit feature (as Veo 3 and Seedance 2.0 currently work on Promptus) or with any additional access requirements carried over from Black Forest Labs' own early-access gating.

Pricing

FLUX 3-specific pricing: pending. Black Forest Labs has not published public pricing for FLUX 3 Video, and Promptus has not yet listed FLUX 3 in its model pricing. We will not publish a number here until it's confirmed — if you see a specific FLUX 3 credit cost quoted anywhere else right now, treat it as unverified.

Promptus generally prices comparable cloud video models:

• Promptus runs on a credit system across its subscription tiers. Simple generations typically use around 1 credit; complex, high-resolution outputs use roughly 3–5 credits. Credits do not expire.
• Artisan — $5/month, 5,000 credits/month, includes standard models and ComfyUI-adjacent tools.
• Designer — $25/month, 25,000 credits/month, unlocks ComfyUI, CosyCloud, custom domains, and additional model access.
• One-time license — $49, unlocks the Promptus desktop app with ComfyUI for unlimited local generation on your own GPU. This covers local, offline generation specifically — cloud-connected models that run on Black Forest Labs' or another provider's servers (which is how FLUX 3 Video will initially work, per BFL's own API-based rollout) are billed separately through the credit system, not covered by the one-time local license.
• As a directional comparison only: Veo 3 video generation on Promptus currently costs 4,500 credits per generation, roughly $5 on the Artisan plan. This is Veo 3's price, not FLUX 3's — we're including it only so you have a realistic sense of scale for what a comparable cloud video model costs on Promptus today.
• Credits can also be earned by sharing idle GPU capacity through Promptus's distributed compute network.

API access

FLUX 3-specific endpoint: Promptus published a FLUX 3 model identifier or endpoint.

Below is Promptus's general API pattern, based on currently live cloud video models (Seedance 2.0), so developers know what shape to expect once FLUX 3 access opens.

Promptus's cloud video generation API follows an asynchronous job pattern:

POST /contents/generations
{
 "model": "<model-id>",
 "mode": "text_to_video" | "image_to_video" | "images_to_video",
 "images": ["url_1", "url_2"],
 "prompt": "description of the desired output"
}


You submit a request with your model ID, generation mode, any reference images, and your prompt. This returns a task ID. You then poll that task ID until the job completes and receive a downloadable output file. This is a fully asynchronous pattern — it does not block your application while generation runs.

If you're building against the Promptus API today, we'd recommend structuring your integration around this same async task-ID pattern now, so switching in the FLUX 3 model ID later requires minimal changes.