GetImageFromSD3byT2I

GetImageFromSD3byT2I is a text-to-image node that doesn't run any model on your machine at all. It calls Stability AI's hosted SD3 API, waits for the picture, saves it to your output folder, and hands the result back to your graph as an IMAGE. For a beginner this is a genuinely interesting option: it's a way to get Stable Diffusion 3 output with zero VRAM requirements, because the actual inference happens on Stability's servers.

There's a second, quieter reason you might want it. The open-weights SD3 Medium (2B) that everyone could download in 2024 was, by community consensus, the broken one - famously terrible at hands, feet, and "girls lying on the grass." The 8B model behind Stability's API was meaningfully better and was never open-released. So this node gives you the good SD3 that was locked behind the paywall. Whether the paywall is worth it is the question, and the answer depends on how much you care about SD3's particular look.

How it works

The node POSTs to https://api.stability.ai/v2beta/stable-image/generate/sd3 with your API key as a Bearer token, your prompt, model, aspect ratio, seed, and output format. On a 200 response it saves the returned bytes to the output directory, loads the file back into a ComfyUI IMAGE tensor, and returns it - so you can preview, upscale, or feed it into another node.

Two behaviors are worth knowing up front:

• sd3-turbo ignores your negative prompt. The node only sends negative_prompt when you select sd3. The README says so plainly, and it matches how the API works. Pick sd3-turbo for speed and cheap credits, sd3 if negative prompting matters to you.
• It's a paid API. Your Stability account needs a credit balance - the README says a minimum of 4 credits - and each call burns some. Author's numbers: sd3 costs 6.5 credits, sd3-turbo costs 4. 1,000 credits runs you about $10. This is real money per image, so it's not a toy for generating hundreds of images.

The inputs that matter

• key - your Stability AI API key (sk-...). This is the whole game: no key, no image.
• positive / negative - your prompts. Both multiline. Remember the turbo/negative caveat above.
• aspect_ratio - pick from 16:9, 1:1, 21:9, 2:3, 3:2, 4:5, 5:4, 9:16, 9:21. Default 1:1.
• model - sd3 or sd3-turbo. Turbo is faster and cheaper but skips the negative prompt.
• format - png or jpeg.
• seed - a standard ComfyUI seed widget with its randomize toggle. Set one for reproducible results, or let it roll for variety.

Output is a single IMAGE - wire it into a Preview or SaveImage node if you want it on screen.

Installing it

Part of the aburahamu/ComfyUI-RequestsPoster pack. ComfyUI Manager → "Install Custom Nodes" → search "requests" → install ComfyUI-RequestPoster → restart. Or manually:

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

The pack's requirements.txt is just requests, tweepy, Pillow, numpy, torch - nothing heavy for this node, and no model files to download because the model lives on Stability's side. Then grab an API key from the Stability AI platform and make sure your account has credits.

Common issues

API error on queue. Check the key first, then the credit balance. The author is honest that what happens when you run out of credits mid-generation is "uninvestigated" - the node just raises whatever error Stability returns. The safest reading: keep an eye on your balance.

Aspect ratio ignored / weird file. Minor quirk from the source: the saved file is always named .png even when you selected jpeg. The bytes are valid either way, but the extension lies, so some viewers may complain. Cosmetic, not fatal.

Your API key is in the PNG. This one matters. Your key lives in the workflow JSON, and ComfyUI embeds that JSON into every saved PNG. Sharing a workflow (or the image itself) hands your Stability key to whoever downloads it - and this node is a pay-per-image API, so that's handing them your credit card. Never share a workflow that has a live key in it.

One honest caveat about the model itself: SD3's whole family history is messy - the licensing drama, the CivitAI bans, the fact that the ecosystem never really formed. By 2026 most people building new stuff are on Flux or newer models. But if you specifically want SD3's output - its strong text rendering, its compositions - this node is the lowest-friction way to sample it, especially if your machine can't run an 8B-class model locally.