ByteDance Seedream V4.5 Edit (Official Edition)
Image-to-image with better instruction-following
- images
- api_config
- image
- request_id
- urls
The image-edit version of Seedream V4.5: feed it images plus a prompt, get an edited image back, all rendered in the cloud through BizyAirPlus's official API route. It's the natural next step after a text-to-image pass - take what you generated and push it the last 10% with an edit prompt.
Editing is where closed flagship models earn their keep, because edit quality is really instruction-following quality, and that's the dimension where Seedream is strongest relative to most open checkpoints. Change the lighting, replace the background, restyle the subject, fix a detail - the V4.5 Edit node handles all of it with the same two-field simplicity as its T2I sibling. Compared to V4 Edit it's the same shape of node with a newer model underneath; compared to the 5.x line it's the settled, dependable middle.
How it works
Output node in BizyAirPlus's cloud layer. With BizyAirPlus ON in the action bar, the node uploads your reference images (the images input is a COMFY_AUTOGROW_V3 socket, accepting single images or lists) and sends them with your prompt to Seedream's edit endpoint. The result returns as a normal IMAGE tensor. Official Edition refers to the API route - the vendor's direct one rather than BizyAir's channel.
The inputs that matter
- images - the reference(s) to edit.
- prompt - what to change. Name the edit ("evening light," "empty the street," "red dress") rather than re-describing the whole picture; that's where edit models respond best.
- size - output dimensions as a string, default
2048x2048.
Plus the shared helpers: api_config (per-request base URL/API key override) and skip_error (placeholder output on failure instead of aborting the workflow).
The outputs
image wires into anything downstream - save it, upscale it, chain another edit on it. urls is the direct link, request_id the support trace ID.
Installing it
In BizyAirPlus. Manager search "BizyAirPlus", or:
cd ComfyUI/custom_nodes
git clone https://github.com/siliconflow/BizyAirPlus.git
cd BizyAirPlus
python -m pip install -r requirements.txt
Restart, toggle BizyAirPlus ON, enter the key from bizyair.ai (stored at ~/.BizyAirPlus/apikey.ini or BIZYAIR_API_KEY). Dependencies are the bizyair-cloudberry and bizytrd pip packages; no models to download.
Gotchas
Button OFF or missing key covers most first-run failures - the ComfyUI console shows the real error. And the reputation note, one last time: BizyAir is legit SiliconFlow software; the malware claims trace to an old Manager auto-install mixup. For the node: keep the size string formatted like 2048x2048, and if an edit keeps missing, trim the prompt to the single most important change. Edits are cheaper to iterate on than full regenerations, but they still bill per request.
Inputs (5)
| Name | Type | Default | Description |
|---|---|---|---|
| images | COMFY_AUTOGROW_V3 | — | |
| prompt | STRING | Prompts | |
| sizeopt | STRING | 2048x2048 | Specifies the width and height pixel values of the generated image |
| api_configopt | BIZYAIR_OPENAPI_CONFIG | 单次请求覆盖base_url和api_key | |
| skip_erroropt | BOOLEAN | false | 开启后遇到错误不中断工作流,输出对应类型的错误占位符 |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| image | IMAGE | — |
| request_id | STRING | — |
| urls | STRING | — |