flux2Dev / sdcpp / createVariant
Flux 2 Dev img2img without the 50GB download
- image
- loras
- api_config
- images
- errors
- workflow_id
- raw_json
The img2img sibling of the Flux 2 Dev cloud node. Same story: the 32B model runs on Civitai's sdcpp fleet, you supply a prompt and a source image, and a re-rendered variant comes back as a ComfyUI IMAGE - with no local VRAM and no 50GB download. If you've ever wanted Flux 2 Dev's comprehension applied to an existing picture rather than dreamed from scratch, this is the node.
That's the honest reason to reach for it. Flux 2 Dev is the best open-weights instruction-comprehension model around, and variant work is where that understanding shows up - describe a real change to a photo ("turn this into a rainy night scene, keep the composition") and the model actually does it, at strengths where SDXL-class models would have collapsed into mush. Cloud billing is the price of admission.
How it works
Standard pack plumbing: submit engine: sdcpp, ecosystem: flux2Dev, operation: createVariant to the Orchestration API with your prompt, image, and strength, long-poll, get the IMAGE out. No model input - the recipe pins Flux 2 Dev server-side.
The inputs that matter
- prompt (required) - the change you want. With Flux 2's text encoder, a specific sentence outperforms tag soup every time.
- image (required,
IMAGE) - the source. Anything in your graph. - strength - the control that matters. Flux 2 Dev's adherence lets you push high: at 0.8 it will restructure the scene faithfully to your prompt, which would be a coin-flip on older models. Start 0.5–0.7 for restyle, 0.8+ for real rewrites.
- steps / cfg_scale - 20 / 1 defaults; CFG 1 is the guidance-distilled intent.
- width / height - 512–2048, so you can re-render at a higher resolution than the input, unlike the Flux.1 variant nodes which cap near 1024.
Install
Pack install, same as always: ComfyUI Manager → "Civitai Comfy Nodes" → Install → restart, or clone https://github.com/civitai/civitai-comfy-nodes.git into custom_nodes, pip install -r civitai-comfy-nodes/requirements.txt (just requests), restart.
Common issues
Credentials: Civitai Auth node into api_config or CIVITAI_API_TOKEN env var (key at civitai.com/user/account); headless needs the env var.
Variant-specific: expect cost to be the loudest surprise. A 32B img2img at 20 steps is one of the priciest per-image jobs in the entire pack - heavier than the Flux.1 nodes, definitely heavier than SDXL. There's no local alternative if you don't have the hardware, but budget accordingly and resist the quantity input (max 4) until you've verified a single result.
One more: because the recipe is fixed server-side, a changed Flux 2 Dev version or setting on Civitai's end changes what you get without warning. That's the early-preview reality the pack README owns up to.
Inputs (16)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | — | |
| image | IMAGE | Either A URL, A DataURL or a Base64 string | |
| output_formatopt | COMBO | 4 options: , jpeg, png, webP | |
| image_metadataopt | STRING | External metadata that will be stored with the image | |
| widthopt | INT | 1024512–2048 | — |
| heightopt | INT | 1024512–2048 | — |
| seedopt | INT | 00–4294967295 | — |
| quantityopt | INT | 11–4 | — |
| cfg_scaleopt | FLOAT | 1.001–20 | — |
| stepsopt | INT | 204–50 | — |
| sample_methodopt | COMBO | 16 options: , euler, heun, dpm2, dpm++2s_a, dpm++2m, +10 | |
| scheduleopt | COMBO | 12 options: , simple, discrete, karras, exponential, ays, +6 | |
| negative_promptopt | STRING | — | |
| lorasopt | CIVITAI_LORAS | — | |
| strengthopt | FLOAT | 0.700–1 | — |
| api_configopt | CIVITAI_CONFIG | Optional Civitai Auth connection; defaults to CIVITAI_API_TOKEN or stored OAuth login. |
Outputs (4)
| Name | Type | Description |
|---|---|---|
| images | IMAGE | — |
| errors | STRING | — |
| workflow_id | STRING | — |
| raw_json | STRING | — |