What is AI Scribble to Image?

Make a rough scribble — even a stick-figure-level doodle — and the AI renders it as a finished image. The scribble guides composition (rough shapes, layout, scene structure); the prompt fills in style and detail.

How is this different from /image/sketch-to-image/?

Sketch-to-image (Canny) wants clean precise line art — your strokes ARE the edges. Scribble-to-image is much more forgiving — uses PiDiNet soft-edge detection that smooths over messy scribbles, missed edges, kid-drawing-quality input. Use sketch for clean linework, scribble for rough pen / mouse / finger sketches.

What kind of input works?

Any drawing — pencil, pen, marker, digital scribble, stick figures, kindergarten-level art. Photos of paper sketches work too. The detector smooths and abstracts your strokes, so quality of linework matters less than for canny.

Which scribble model powers it?

PiDiNet detector (with HEDdetector fallback) extracts soft-edge maps from your input; xinsir/controlnet-scribble-sdxl-1.0 ControlNet then conditions SDXL on those edges. Best-in-class for messy / loose / abstract inputs.

Yes — ~1,200 tokens per render (1,000 base + 20% ControlNet surcharge). 5,000 free tokens/day for signed-in users covers about 4 scribbles daily.

How do I prompt it well?

Be SPECIFIC about subject + style. The scribble fixes composition, the prompt fills the rest. "A red dragon breathing fire on a castle, fantasy oil painting, dramatic lighting" beats "a dragon".

Can I tune how strictly the scribble is followed?

ControlNet strength slider (default 0.7). 1.0 = strict (output looks like a refined version of your exact scribble shape). 0.4 = loose (the prompt + scribble negotiate; output is a cleaner re-imagination). Lower it if your scribble is just a vibe.

Storyboarding (rough panel sketches → final art), kids' drawings rendered as paintings, quick concept sketches turned into pitch art, brainstorming visual ideas without the skill barrier of clean drawing.

How does it compare to ControlNet QR / Tile / Brightness alternatives?

Different control surfaces. Scribble = freehand structural guidance. QR / Tile / Brightness optimize for different things (hidden patterns, photo upscale, lighting). Scribble is the right pick for "I drew something, render it."

What output sizes are supported?

512×512 default. Up to 1024×1024 on standard SDXL aspect ratios. Larger sizes consume more VRAM and tokens.

Are the scribble uploads stored?

Processed, scribble-extracted, then discarded. Only the final render + prompt are kept on /account/?tab=history. Never used for training. /privacy/.

Yes — POST multipart to /v1/image/generate/ with model=sdxl, prompt, control_image (file), control_type=scribble. Bearer auth, 10K free monthly tokens. /api/ has curl examples.

Снимок (контрольная сеть)

Загрузите рисунок, линейный рисунок или любое изображение — мы извлекаем его контур (наймите кромку) и используем его для контроля генерируемого изображения.

Контрольный снимок (рисунок/набросок/фотография)

Мы запустили обнаружение края Кэнни, а более чистые наброски дают более чистую контрольную систему.

Быстро

Контрольная сила

Расслабься. 0.7 Усилить

Соотношение спектаклей

~1,200 tokens (SDXL × 1.2 ControlNet)

Снимок (контрольная сеть)

Результат

Получить 10 000 бесплатных токенов

Подождите — найдите 10 тысяч бесплатных токенов!

Хочешь еще?