Bytedance Dreamina Edit
Edit an image with a prompt, no mask painting
- client
- image
The companion piece to the Dreamina Text-to-Image node, and arguably the more useful one: give it a URL of an existing image plus a text prompt, and it returns an edited IMAGE tensor. "Change the background to a rainy street," "turn this product shot into a studio scene" - the model handles the semantic edit, which beats hand-painting a mask for most content changes.
One honest warning up front, because it's the first thing that surprises people: the input is image_url, a plain STRING - not an IMAGE socket. There's no tensor input on this node. The source image must be reachable at a public HTTPS URL. That's the one real design quirk here, and it shapes how you wire the thing.
How it works
Like its sibling, it's a WaveSpeed API call. The node hands Dreamina the image URL and the prompt, polls the job, and converts the result to a tensor on the image output. Because the source is a URL, the natural pairing is with other cloud nodes: generate with Dreamina Text-to-Image, then grab the URL... except that node only outputs a tensor. The practical sources are a URL stored in a text node, a file you've hosted, or an output URL from another API node that exposes one.
Inputs
- prompt - required, the modification instructions. "Describe the desired image modifications," per the tooltip.
- image_url - required. The public URL of the image to edit.
- width / height - both default 1328, range 512–2048. The edit output size, not a crop.
- seed - -1 randomizes (fresh billed edit each run); fixed value = cache control.
- enable_sync_mode - off. Wait-for-completion toggle; you can leave it.
- enable_base64_output - off. Leave it.
The output is a single IMAGE tensor - into a Preview/Save node or onward in your chain.
Install and gotchas
Part of the ERPK Collection:
cd ComfyUI/custom_nodes
git clone https://github.com/eRepublik-Labs/comfyui-nodes-erpk.git erpk
cd erpk && pip install -r requirements.txt
or via ComfyUI Manager → search erpk. Add a WaveSpeed API key (wavespeed.ai) under Settings > ERPK > API Keys - every edit is a paid API call.
The failure modes are mostly URL-shaped. Private/localhost URLs fail because WaveSpeed's servers can't reach them - you need something publicly fetchable, so the "load a file from your hard drive" instinct doesn't work here. If the edit comes back looking like it never saw the source image, the URL probably 404'd server-side or was a redirect the API didn't follow. And keep in mind this node is stuck on the Dreamina V3.0 edit model under the hood - no V3.1 variant is offered here like the t2i node has, so if you need the newest edit model, that's a different node in the same pack.
Inputs (8)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | Text description of the desired image modifications | |
| image_url | STRING | URL of the image to edit | |
| clientopt | WAVESPEED_AI_API_CLIENT | WaveSpeed API client (optional if API key is configured in Settings) | |
| widthopt | INT | 1328512–2048 | Image width (512 to 2048) |
| heightopt | INT | 1328512–2048 | Image height (512 to 2048) |
| seedopt | INT | -1-1–2147483647 | Random seed for reproducibility (-1 for random) |
| enable_sync_modeopt | BOOLEAN | false | Wait for completion before returning response |
| enable_base64_outputopt | BOOLEAN | false | Return BASE64-encoded output instead of URL |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| image | IMAGE | — |