Flux-2-pro
Multi-image editing with color you can aim
- input_images
- output
Flux 2 is Black Forest Labs' current flagship, and the interesting part for most people is that you can't have it locally - the useful tier is closed, and the open ecosystem around it lost the race on exactly that point. So if you want Flux 2 Pro's multi-image editing and its unusually precise color control inside ComfyUI, an API wrapper is the only door. This node is that door, pointed at the pack's single gateway. One key, and the whole FLUX family is in your graph.
How it works
You send a prompt (and optionally one or more reference input_images), the node base64-encodes everything and POSTs to the gateway under the model id flux2. The server runs Flux 2 Pro and returns images, which the node downloads into IMAGE tensors. The author's own framing: Flux 2 supports Chinese and English in and out, does multi-image editing well, and - its differentiator against Nano Banana - "supports precise color editing." That last bit is the genuinely interesting workflow: feed a product photo plus a color reference and ask for a recolored variant, and Flux 2 tends to nail the hue instead of approximating it.
The inputs that matter
prompt(STRING) - the edit instruction. Multiline, so you can be verbose.is_translation(BOOLEAN, default false) - the pack's standard Chinese-input toggle. If your Chinese prompt comes back under-served, flip this on; it pre-translates before the model sees it.aspect_ratio(auto, 16:9, 4:3, 1:1, 3:4, 9:16) - leave onautoand the model keeps your reference's framing, which is usually what you want for edits.custom_size+width/height(1024–2048) - turncustom_sizeon to override the aspect ratio with exact pixels.seed(INT) - reproducibility.
Optional: input_images (IMAGE - a batch, for multi-image edits). One output, output (IMAGE).
Installing it
This is one node in comfyui-MJAPI-party. ComfyUI Manager → search "mojieapi_party" → Install → restart, or:
cd ComfyUI/custom_nodes
git clone https://github.com/MoJIeAIGC/comfyui-MJAPI-party.git
Register at mojieaigc.com and set your key in config.ini (leave BASE_URL = https://www.mojieaigc.com/v1/completions as shipped):
[API]
KEY = your_api_key
BASE_URL = https://www.mojieaigc.com/v1/completions
Restart. No downloads; the standard numpy/torch/pillow/requests dependencies only.
Common issues
- It's the pro tier, so it's not the cheap tier. Flux 2 Pro per-image pricing sits at the premium end, and with the whole reseller layer marking it up, watch your balance. For pure prompt-to-image at scale, Seedream or Nano in this same pack tend to be friendlier to the wallet.
- Chinese prompt came out wrong. That's what
is_translationexists for - try it before assuming the model failed. - Color edits that drift. If the recolor "helps" but picks its own palette, give the model an explicit target ("exactly the RGB of the reference object") and feed the reference in
input_images. Vague color words are where Flux 2 still approximates. - Error image with text - the pack's standard failure face; usually key, balance, or a refused edit (moderation follows BFL's filters, and nothing in the node bypasses it).
Inputs (8)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | A beautiful sunset | — |
| is_translation | BOOLEAN | false | — |
| aspect_ratio | COMBO | auto | 6 options: auto, 16:9, 4:3, 1:1, 3:4, 9:16 |
| custom_size | BOOLEAN | false | — |
| width | INT | 10241024–2048 | — |
| height | INT | 10241024–2048 | — |
| seed | INT | 0 | — |
| input_imagesopt | IMAGE | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| output | IMAGE | — |