Question 1

What is the Free AI ControlNet generator?

Accepted Answer

A single tool that exposes all 12 conditioning types from the ControlNet-Union SDXL ProMax model — canny, pose, depth, scribble, lineart, anime-lineart, MLSD, HED, soft-edge, normal, segmentation, and tile. Pick a conditioning type, drop a reference image, write a prompt, and SDXL renders a new image that follows the structure of your reference.

Question 2

How is ControlNet different from img2img or image-to-image?

Accepted Answer

img2img re-paints over the input directly — colors, edges, AND geometry mix with the prompt. ControlNet throws the colors away and keeps just the chosen structural signal (lines, pose skeleton, depth map, etc.). That lets you radically swap content while keeping composition rock solid. Far stronger structural control than img2img.

Question 3

Which conditioning type should I pick?

Accepted Answer

Canny / lineart for clean linework input. Anime-lineart for anime-style line input. Scribble / soft-edge / HED for rough sketches and doodles. Pose to copy a body position from a photo. Depth to preserve scene geometry / 3D layout. MLSD to preserve straight lines (architecture / interiors). Normal to preserve surface orientation and volume. Segmentation to preserve regions. Tile to refine or upscale variations of an existing image.

Question 4

Why one model for all 12 types?

Accepted Answer

ControlNet-Union SDXL ProMax (xinsir, Apache 2.0) packs all 12 conditioning networks into a single 2.5 GB weight. Older deployments downloaded a separate ~2.5 GB weight per type — switching between canny and pose meant a cold-start. The union model loads once and stays warm, so every conditioning type is sub-second after the first call.

Question 5

Is it really free?

Accepted Answer

Yes. ~1,200 tokens per render (1,000 base SDXL + 20% ControlNet conditioning surcharge). Signed-in users get 30,000 free tokens daily — about 25 conditioned renders per day at no cost. Anonymous: 2,500 tokens/day (~2 renders).

Question 6

Can I tune how strictly the AI follows the conditioning?

Accepted Answer

Yes — the Control strength slider (default 0.7) determines how strictly the output follows your reference. 1.0 = strict (output looks like a re-render of your reference). 0.4 = loose (prompt has more freedom). Lower it for creative variation, raise it when fidelity matters.

Question 7

What output sizes are supported?

Accepted Answer

512×512 default. SDXL standard ratios — 768×1024 portrait, 1024×768 landscape, 1024×1024 square — all work. Larger outputs consume more VRAM and tokens; the H200 supports up to 1024×1024 comfortably.

Question 8

Are the uploaded reference images stored?

Accepted Answer

Reference images are processed immediately, conditioning is extracted, then the reference file is deleted. Only the prompt + final render remain on /account/?tab=history. Never used for training. /privacy/ for the full policy.

Question 9

What license is the model under? Can I use the output commercially?

Accepted Answer

ControlNet-Union SDXL ProMax is released under Apache 2.0 — fully permissive, including commercial use. SDXL base is OpenRAIL++. Both allow commercial use; your generated images are yours to use commercially with no royalties.

Question 10

How does this compare to ComfyUI or Automatic1111 with ControlNet?

Accepted Answer

Same model, same quality, same conditioning signals. ComfyUI and A1111 require a local GPU with 12+ GB VRAM plus setup. We run it on shared infrastructure with a generous free pool — no install, no GPU needed.

Question 11

Why does the first render after a cold start take longer?

Accepted Answer

First call downloads the Union weight (~2.5 GB) into the GPU cache and warms the SDXL pipeline. Expect 30-60 seconds on the very first request after a deploy or LRU eviction. Subsequent calls under typical load return in 4-7 seconds.

Question 12

Is there an API?

Accepted Answer

Yes — POST multipart to /v1/image/generate/ with model=sdxl (or model=controlnet-union-sdxl-promax), prompt, control_image (file), control_type=<one of: canny, pose, depth, scribble, lineart, anime-lineart, mlsd, hed, soft-edge, normal, segmentation, tile>, optional control_strength (0.1-1.5). Bearer auth, 10K free tokens/month. /api/ has curl examples.

ControlNet — 12 conditioning types in one tool

Result

ControlNet — 12 conditioning types in one tool

Result

Get 10,000 Free Tokens

Wait — Get 10K Free Tokens!

Want more?