FalFluxI2IAPI

FalFluxI2IAPI is the image-to-image half of the pack's Flux story. You feed it an image tensor, a prompt, and a strength, and it runs fal-ai/flux/dev/image-to-image in the cloud, returning a re-imagined IMAGE. For anyone on a weak GPU this is the practical way to do Flux img2img - no 12B transformer, no T5, no VAE decode/encode round trips eating your VRAM. Load an image, type what should change, get a result.

How it works

The node uploads your image to fal, calls the Flux dev i2i endpoint with strength, steps, cfg, and seed, then downloads the output. Two things it does on the way in are worth knowing:

• It auto-downsizes your image to 1024px on the longest edge before uploading, unless you flip no_downscale on. This is a cost guardrail - fal bills by the image, and oversized inputs just burn pixels. It's the right default; leave it alone unless you genuinely need full-res input.
• Flux i2i is sensitive. The pack author's guidance from the field: strength between 0.90 and 0.95 works well for realism, flatter/stylized styles tolerate it less. This matches the general Flux wisdom that the model is a lot more delicate about denoise than SDXL was.

There's no width/height input on this node - the output takes its size from the input image.

The inputs that matter

• image - the IMAGE you're editing. Wire it from a Load Image node.
• prompt - describe the change. Flux's T5 encoder reads full sentences, so "turn this into a rainy night street" beats tag soup.
• strength - 0.01 to 1.0, default 0.9. How much of the original survives. Think of it as the denoise dial.
• steps - default 25, right for dev. Fewer steps = faster = cheaper.
• cfg - Flux guidance, default 3.5.
• no_downscale - false by default; true skips the 1024px cap on upload.
• image_count - 1 to 4 variations, returned as one batched IMAGE.

Output is a single IMAGE tensor. If you asked for several, use the pack's SplitImages node to break the batch apart.

Install

Same as every node in this pack - ComfyUI Manager, search "ComfyUI-Cloud-APIs", or:

cd ComfyUI/custom_nodes
git clone https://github.com/BetaDoggo/ComfyUI-Cloud-APIs

Restart, put a fal key in a .txt file in ComfyUI-Cloud-APIs/keys/, and top up at fal.ai/dashboard/billing. No models to download - that's the deal.

Troubleshooting

• "Works but not the img" - the classic i2i complaint (literally the first bug report in the pack's launch thread). It's a strength/settings issue, not a bug: Flux i2i is finicky. Start at 0.9, adjust, don't expect SDXL-style 0.5 strengths to behave.
• Output resolution feels low - check whether no_downscale is off and your source was bigger than 1024; the downscale happened on purpose, to save you money.
• Cloud costs - every iteration is a billed generation. Preview small, iterate the prompt, then spend.
• Safety filters - dev/schnell ship with safety checkers disabled in this pack; pro runs at max tolerance. Fine for most work, but don't assume any API tier is a content guarantee.

Pack's archived, node keeps working because it's a thin HTTP wrapper. If fal changes the endpoint, fork it.