Upscale

Drop or pick an image; it comes back upscaled, ready to download.

Which engine runs, and what the local one needs

Cloud (Topaz, via Kie) — the default. Runs on Kie's hardware, so it works from any machine including a laptop with no discrete GPU, and from the hosted studio. Costs Kie credits per image.

Local (Upscayl / Real-ESRGAN) — free and offline, but it runs on your GPU and is only available where the Upscayl binary is installed. Minimum realistic specs:

• Discrete GPU with Vulkan support — integrated graphics will crawl or fail
• ~8 GB VRAM for a 4× on a 2K image. 4 GB cards (e.g. GTX 950M) run out of memory at 4×/2K — drop to 2×, or use the cloud engine
• Expect minutes, not seconds, on modest hardware

Not installed on this server? Then the cloud engine is used automatically — nothing to configure.

Recycle bin

Soft-deleted images. Restore them to the pipeline or delete permanently.

Nothing here.

Agents — generate images via the n8n webhook

One HTTP POST returns a hosted image URL. Give agents this endpoint and the recipe presets below. Full reference lives in the skill on GitHub.

↗ Open the anima-n8n-webhook skill on GitHub

Endpoint

POST https://n8n.divergence.systems/webhook/anima/generate
Content-Type: application/json

Request body

{
  "prompt": "who + doing what + where — one concrete sentence, no style words",
  "model": "nano-banana-pro",
  "aspect_ratio": "3:2",
  "resolution": "2K",
  "controls": "optional: lens + light + mood (use a recipe below)"
}

aspect_ratio: 1:1 · 3:2 · 2:3 · 16:9 · 9:16. resolution: 1K · 2K. The realism floor (photoreal scaffold + anti-AI-tell negatives) is added automatically, so do not write "photorealistic" or "4K" yourself. Provider defaults to Kie; do not use OpenAI.

Image models (pass as model)

Response

{ "ok": true, "url": "https://res.cloudinary.com/.../anima/xxxx.jpg", "engine": "Kie" }

Recipe presets (aspect_ratio + controls)

Copy-paste example

curl -s -X POST https://n8n.divergence.systems/webhook/anima/generate \
  -H "Content-Type: application/json" \
  -d '{"prompt":"A friendly migration consultant advising a relaxed client across a desk, Thailand visa documents on the table","model":"nano-banana-pro","aspect_ratio":"3:2","resolution":"2K"}'

Copy example

Prompt library

Save your best prompts to reuse. "New prompt" grabs whatever is in the Studio composer; "Use in Studio" loads one back in.

+ New prompt

Service keys

Which keys this studio is actually running on, and whether they work. A key that is set and a key that works are different things, so each one is tested live against its provider. Values are masked and never sent back in full.

Changing a key needs the fleet token, not the studio token — the studio token is scoped to generating images, and letting it rewrite the keys would turn it into control of everything the studio spends money through. Overrides are stored on the server volume and survive a rebuild; clearing one falls back to the deployed environment.

Logs

Recent generation events. Failures include the per-lane error so you can see exactly what went wrong (and which engine flaked).

Clear logs