FLUX 3 ComfyUI
August 28, 2026 · View on GitHub
ComfyUI custom nodes for the FLUX 3 API, delivered through Muapi.
The pack follows the focused Muapi workflow pattern used by MiniMax H3 ComfyUI: add one API Key node, connect it to a generation node, then save or preview the returned media.
The FLUX 3 video API is live on Muapi. Text-to-Video, Text-to-Video Draft, Image-to-Video, Start-End-to-Video, Video Extend, Video Extend Draft, and Video Upscaler all work today with a Muapi API key. Text-to-Image, Image-to-Image, and Dev remain coming soon/early access upstream — those three nodes are wired to the exact API contract so the same workflows will start working as soon as Muapi enables them.
📺 FLUX 3 API: Image & Video Generation from One Multimodal Model →
Included nodes
| Node | What it does | Status |
|---|---|---|
| FLUX 3 API Key | Stores one Muapi API key for the workflow. | — |
| FLUX 3 Text-to-Image | Generates a 1k, 2k, or 4k image from a prompt. | Coming soon |
| FLUX 3 Dev | Uses the faster, lower-cost FLUX 3 Dev image route. | Coming soon |
| FLUX 3 Image-to-Image | Edits an image using up to four references and an instruction prompt. | Coming soon |
| FLUX 3 Text-to-Video | Generates a 5–20 second video at 720p or 1080p, with optional native audio. | Live |
| FLUX 3 Text-to-Video Draft | Fast, lower-cost draft mode of Text-to-Video for prompt iteration. | Live |
| FLUX 3 Image-to-Video | Animates one start frame into a video, with optional native audio. | Live |
| FLUX 3 Start-End-to-Video | Generates a transition between a start and end keyframe. | Live |
| FLUX 3 Video Extend | Continues an existing clip with new prompt-guided motion. | Live |
| FLUX 3 Video Extend Draft | Fast, lower-cost draft mode of Video Extend. | Live |
| FLUX 3 Video Upscaler | Upscales any video beyond its native resolution (Precise or Creative tier). | Live |
| FLUX 3 Save Image | Downloads a generated image into ComfyUI/output and returns an IMAGE. | — |
| FLUX 3 Save Video | Downloads a generated video into ComfyUI/output and returns decoded frames. | — |
Installation
Copy this directory into ComfyUI/custom_nodes/flux-3-comfyui, then install the dependencies in the same Python environment ComfyUI uses:
pip install -r ComfyUI/custom_nodes/flux-3-comfyui/requirements.txt
Restart ComfyUI. The example workflows can be loaded with File → Load.
Included examples:
FLUX3_T2I_Example.jsonFLUX3_I2I_Example.jsonFLUX3_T2V_Example.jsonFLUX3_I2V_Example.jsonFLUX3_StartEndToVideo_Example.jsonFLUX3_VideoExtend_Example.json
API key
Use the 🔑 FLUX 3 API Key node and connect its output to a generation node. Alternatively, leave the node field empty and configure one of these:
export MUAPI_API_KEY="your_muapi_key"
or:
muapi auth configure --api-key YOUR_KEY
The nodes also read ~/.muapi/config.json when it contains an api_key field. Keys are only used for requests and are not written into generated workflow files unless you explicitly save them in a node field.
Quick workflows
Text-to-image
- Add 🔑 FLUX 3 API Key.
- Add 🖼️ FLUX 3 Text-to-Image or ⚡ FLUX 3 Dev.
- Choose an aspect ratio and
1k,2k, or4kresolution. - Connect
imageto 🖼️ FLUX 3 Save Image or ComfyUI's Preview Image node.
Image-to-image
Connect up to four Load Image → IMAGE outputs to the image_1 … image_4 inputs on 🖌️ FLUX 3 Image-to-Image. Each slot also has an image_url_N field for a hosted URL or a local file path. Connected IMAGE inputs take precedence over their URL fields.
Use an instruction-style prompt, for example:
Replace the background with a warm, softly lit studio while keeping the subject's identity, pose, and clothing unchanged.
Text-to-video
Connect 🔑 FLUX 3 API Key → 🎬 FLUX 3 Text-to-Video, choose an aspect ratio, 720p or 1080p, a duration from 5–20 seconds, and whether to generate synchronized native audio. Connect video_url to 🎬 FLUX 3 Save Video and first_frame to Preview Image. Use 🎬 FLUX 3 Text-to-Video Draft for cheap, fast iteration before a final render — it skips the resolution control and bills a flat rate.
Image-to-video
Connect Load Image → IMAGE to 🎬 FLUX 3 Image-to-Video, or provide a hosted URL/local path in image_url. The node uploads the image when needed and sends it as the single image_url required by the API.
Start-end-to-video
Connect two Load Image → IMAGE outputs to start_image and end_image on 🎬 FLUX 3 Start-End-to-Video (or use the start_image_url / end_image_url fields), and describe the transition in the prompt.
Video extend
Provide a source clip URL or local path (under 50MB, under 15s) in video_url on ➡️ FLUX 3 Video Extend, and describe the continuation in the prompt. Use ➡️ FLUX 3 Video Extend Draft for fast, low-cost continuity testing — it skips the resolution control and bills a flat rate.
Video upscaler
Provide a source video URL or local path in video_url on 🔎 FLUX 3 Video Upscaler, set upscale_factor$ (1–4 \times ), \text{and} \text{choose} $creativity: 0 stays closer to the source (Precise, $1.43/run), 1 reconstructs more detail (Creative, $2.00/run).
API contract
| Workflow | Endpoint | Required fields | Status |
|---|---|---|---|
| Text-to-image | POST /api/v1/flux-3-text-to-image | prompt | Coming soon |
| Image-to-image | POST /api/v1/flux-3-image-to-image | prompt, images_list (up to 4) | Coming soon |
| FLUX 3 Dev | POST /api/v1/flux-3-dev | prompt | Coming soon |
| Text-to-video | POST /api/v1/flux-3-text-to-video | prompt | Live — $0.25/s (720p), $0.42/s (1080p) |
| Text-to-video draft | POST /api/v1/flux-3-text-to-video-draft | prompt | Live — $0.09/s flat |
| Image-to-video | POST /api/v1/flux-3-image-to-video | prompt, image_url | Live — $0.25/s (720p), $0.42/s (1080p) |
| Start-end-to-video | POST /api/v1/flux-3-start-end-to-video | prompt, image_url, end_image_url | Live — $0.25/s (720p), $0.42/s (1080p) |
| Video extend | POST /api/v1/flux-3-video-extend | prompt, video_url | Live — $0.25/s (720p), $0.42/s (1080p) |
| Video extend draft | POST /api/v1/flux-3-video-extend-draft | prompt, video_url | Live — $0.09/s flat |
| Video upscaler | POST /api/v1/flux-3-video-upscaler | video_url | Live — $1.43/run (Precise), $2.00/run (Creative) |
| Poll task | GET /api/v1/predictions/{request_id}/result | — | — |
| Upload local file | POST /api/v1/upload_file | multipart file | — |
Image routes send aspect_ratio and resolution. Non-draft video routes send aspect_ratio, resolution, duration, and generate_audio; draft video routes send the same minus resolution. Duration is billed rounded up to the next whole second, capped at 20 seconds; audio generation is included at no extra charge. All requests use the x-api-key header and generation is asynchronous.
For local testing against a compatible Muapi deployment, set FLUX_3_API_BASE_URL. Polling can be tuned with FLUX_3_POLL_INTERVAL and FLUX_3_MAX_WAIT.
Requirements
Python 3.8+, requests, Pillow, numpy, and opencv-python. ComfyUI supplies the installed PyTorch runtime used by IMAGE tensors. For video playback inside ComfyUI, install ComfyUI-VideoHelperSuite.
Related Projects
- MiniMax H3 ComfyUI — the focused reference pack for this repository's node layout.
- Flux-3-Dev-API — Python wrapper for the FLUX 3 family through Muapi.
- flux-3-video-edit — Python SDK for FLUX 3's instruction-driven video editing mode (coming soon)
- flux-3-video-upscaler — Python SDK for upscaling FLUX 3 (or any) video output beyond its native resolution (coming soon)
- flux-3-omni — Python SDK for FLUX 3's multi-reference Omni Reference mode (coming soon)
- Muapi API docs — FLUX 3 capabilities and availability.
- Muapi ComfyUI integration — the full multi-model node pack.
License
MIT