GetImageFromSD3byI2I

GetImageFromSD3byI2I is the image-to-image sibling of the pack's text-to-image node. Instead of generating from scratch, it takes an image you already have, sends it to Stability AI's hosted SD3 API, and gets back a version re-rendered according to your prompt. It's a way to push your own image through SD3's particular look - its compositions, its text rendering - without ever running the model on your machine.

Where this earns its keep: you've got an image you like, but you want it restyled, re-composed, or pushed a bit closer to a concept. Or you want to see what SD3 does to something you generated locally. That's exactly the image-to-image use case, and it's the same cloud SD3 the GetImageFromSD3byT2I node hits - including the 8B API model that was never released as open weights. Same caveats, too: it's a paid Stability API, and it doesn't care about your VRAM because nothing runs locally.

How it works

The node does a short dance before it ever talks to the API:

1. It checks your input image is at least 64×64 px - anything smaller raises an error.
2. It saves the input image to your output folder (under the SD3I2I prefix), so you get a copy of what you sent.
3. It POSTs to https://api.stability.ai/v2beta/stable-image/generate/sd3 with mode: "image-to-image", your prompt, a strength value, model, seed, and format.
4. On success it saves the returned image and hands it back as an IMAGE output.

The strength slider is the heart of it. The README's framing is honest and useful: 0 returns your input image basically unchanged, 1 is the same as generating from nothing, and things start changing noticeably around 0.6. Think of it as denoise - low strength means subtle restyle, high strength means the prompt takes over.

Same negative-prompt rule as the T2I node: sd3-turbo ignores it, so the node only sends negative_prompt when you select the plain sd3 model. And same pricing: sd3 burns 6.5 credits per call, sd3-turbo burns 4, with a minimum balance of 4 credits and roughly $10 per 1,000 credits.

The inputs that matter

• images - required IMAGE input. This is what gets re-rendered.
• key - your Stability API key.
• positive / negative - your prompts. The negative only bites on sd3, not sd3-turbo.
• strength - 0 to 1, default 0.5. The single most important dial here; start around 0.5 and walk it up.
• model, format, seed - same as the T2I node: sd3 vs sd3-turbo, png vs jpeg, and the seed widget for reproducibility.
• execute - master switch, default on. Flip it off to keep the node in the graph without spending credits.

Output is a single IMAGE. Unlike the other poster nodes, this one isn't a dead end - you can preview, upscale, or chain the result onward.

Installing it

Same pack as the rest: aburahamu/ComfyUI-RequestsPoster. ComfyUI Manager → "Install Custom Nodes" → search "requests" → install ComfyUI-RequestPoster → restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/aburahamu/ComfyUI-RequestsPoster.git

No model downloads - the model is on Stability's side - and the pack's dependencies (requests, tweepy, Pillow, numpy, torch) are all light and mostly preinstalled. You need a Stability API key with credit balance, nothing more.

Common issues

"画像サイズエラー" or similar error on queue. That's the node's 64px minimum-size check failing - the message is in Japanese since the author is Japanese. Upscale your input above 64×64 and it goes away.

Image comes back basically identical. Your strength is too low. Crank it toward 0.6–0.8 to actually see the re-render do something.

API error / nothing returned. Same playbook as the T2I node: check the key, check the credit balance. The author's README notes the "credits exhausted" behavior is uninvestigated - the node just surfaces whatever error Stability returns, so you'll want to watch your balance.

Input image doubled in your output folder. Not a bug - step 2 above always saves a copy of what you sent before uploading it. If you're feeding this from a generator, your outputs will contain both the source and the result.

And the big one, repeated because it matters: your key ends up in the workflow JSON, which ComfyUI bakes into every saved PNG. This node spends real money per call. Sharing a workflow with a live key in it is handing your Stability credit card to whoever opens the file. Paste the key in, generate, then scrub it before you share anything.

Want the T2I flavor instead? Same pack, sibling node, no image input, just prompt-to-picture. Pick I2I when you have a starting image worth pushing.