mageflow / 4b-edit-turbo / editImage
Four-step instruction editing you iterate on
- images
- loras
- diffusion_model
- api_config
- images
- errors
- workflow_id
- raw_json
This is the draft-machine version of MageFlow's edit model. The 4B-edit-turbo variant does instruction edits in 4 steps at CFG 1.0 - which, on Civitai's cloud, means it's the cheap, fast, iterate-until-it's-right node in the MageFlow family. Same model family as the 30-step edit, distilled down to the speed that makes trial-and-error editing affordable per job.
Reach for this one when you're sketching: try "make it night", "move the subject right", "change the mood" in quick succession, find the keeper, and only then spend the bigger Buzz on the full-fidelity edit node. It's the classic turbo workflow - draft fast, finish slow - except the "GPU" is Civitai's fleet.
How it works
Same civitai-comfy-nodes machinery as every recipe node: submit prompt + image to the Civitai Orchestration API, run MageFlow 4B-edit-turbo on Civitai's Comfy engine, long-poll, return an IMAGE tensor. The turbo distillation is why 4 steps is the default instead of 30 - guidance is effectively disabled (CFG 1.0), so leave those values alone unless you know what you're doing.
The inputs that matter
- prompt and images - both required. Source image plus a plain-language edit instruction.
- steps / cfg_scale - 4 / 1.0 defaults. This is a distilled model; cranking CFG fights the distillation and usually makes output worse, not better.
- diffusion_model (
CIVITAI_AIR) and loras (CIVITAI_LORAS) - optional typed sockets for Civitai Model Selector / LoRA Selector nodes, if you want a finetune or LoRA stack on the cloud job. - sampler / scheduler / seed / width / height / quantity (max 4) - standard.
Outputs
images → Preview/Save. errors, workflow_id, raw_json are diagnostics; raw_json has the full job result.
Installing and authenticating
Pack-wide install:
cd ComfyUI/custom_nodes
git clone https://github.com/civitai/civitai-comfy-nodes.git
pip install -r civitai-comfy-nodes/requirements.txt
or ComfyUI Manager → "Civitai Comfy Nodes" → Install → restart. Dependencies: requests, python-socketio[client].
Auth: Civitai Auth node → CIVITAI_API_TOKEN → stored key/OAuth → browser login. Headless:
export CIVITAI_API_TOKEN=your_token
Where people get burned
- Turbo means turbo quality. 4 steps is fast and cheap, but detail and coherence give way to the non-turbo edit. Don't judge MageFlow's editing ability from turbo drafts alone.
- CFG at 1.0 is a feature. Newcomers see
cfg_scaleand immediately set it to 7. For this node that's a mistake - it's a guidance-distilled model, not SDXL. - Edits re-emit the whole frame, so unedited areas drift slightly and errors compound over chains. Keep the source clean.
- Per-job billing. It's the cheap node in the family, but "cheap" still isn't free - a long batch of drafts is a long batch of Buzz.
- Your prompt and image go to Civitai's servers. Same deal as every cloud node in this pack.
Inputs (16)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | — | |
| steps | INT | 41–150 | — |
| cfg_scale | FLOAT | 1.000–30 | — |
| quantity | INT | 11–4 | — |
| images | IMAGE | — | |
| output_formatopt | COMBO | 4 options: , jpeg, png, webP | |
| image_metadataopt | STRING | External metadata that will be stored with the image | |
| negative_promptopt | STRING | — | |
| sampleropt | COMBO | 32 options: , euler, euler_ancestral, euler_cfg_pp, euler_ancestral_cfg_pp, heun, +26 | |
| scheduleropt | COMBO | 8 options: , normal, karras, exponential, sgm_uniform, simple, +2 | |
| seedopt | INT | 00–4294967295 | — |
| lorasopt | CIVITAI_LORAS | — | |
| diffusion_modelopt | CIVITAI_AIR | — | |
| widthopt | INT | 1024512–2048 | — |
| heightopt | INT | 1024512–2048 | — |
| 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 | — |