Image Generation

Generate icons, diagrams, and artwork with local Stable Diffusion, LLM-driven SVG pipelines, and Python/Pillow code — then save your settings as reusable templates.

Get Started See Features

Built-in pipelines

Stable Diffusion txt2img

Direct text-to-image generation via a local Forge/A1111 WebUI instance. Control model, resolution, steps, CFG scale, sampler, seed, and negative prompts.

SD Refined

LLM expands your brief description into a detailed SD prompt with quality tags, generates the image, then a vision model validates the result against your original request.

SD img2img

Transform existing images with Stable Diffusion. Paste, drag, or upload a source image, adjust denoising strength, and guide the transformation with a text prompt.

SD img2img Refined

Combines LLM prompt refinement with img2img generation and vision validation for iterative image-to-image workflows.

SVG Icon

LLM generates clean, minimal SVG icons from a text description, validates the markup, then refines with feedback for up to 2 iterations.

SVG Diagram

LLM generates SVG diagrams (flowcharts, architecture, sequences) with validation against criteria.

Code Image

LLM writes Python/Pillow code to generate an image programmatically — runs the code via the executor service, validates output, and refines.

HTML Render

LLM generates HTML/CSS, rendered to a pixel-perfect image via a headless browser.

Generate templates

Save and reuse your generation settings

Save from anywhere

Configure a pipeline with the perfect prompt, model, resolution, sampler, and steps — then save it as a named template with one click. You can also save templates directly from any generated image's detail view.

For img2img workflows, templates store the source image as a referenced asset so it's loaded automatically when the template is applied.

Apply and tweak

Select a template from the Generate modal dropdown and every field pre-fills — pipeline, prompt, negative prompt, model, size, sampler, seed, steps, CFG, and source image. Change what you need and generate.

Templates appear as cards in the sidebar of the Images page. Click any card to open a full editor with the same two-column layout as the Generate modal.

Full editing

Every parameter is editable from the template detail modal — name, prompt, negative prompt, model, size, sampler, seed, steps, CFG scale, denoising strength, and source image. Changes are saved via the REST API.

REST API

GET /project/generate-templates

POST /project/generate-templates

PATCH /project/generate-templates/{id}

DELETE /project/generate-templates/{id}

Stable Diffusion integration

Connect to your local Forge or A1111 WebUI

Local GPU, your models

Connects to any A1111/Forge WebUI via REST API. Run on your own GPU — SDXL, SD 1.5, or any checkpoint you have installed.

Full parameter control

Model, resolution, steps (1–50), CFG scale, sampler (Euler a, DPM++ 2M Karras, DDIM, etc.), seed, negative prompt, and denoising strength for img2img.

img2img with source images

Paste, drag-and-drop, or upload a source image. Adjust denoising strength to control how much the AI transforms it. Source images are preserved in templates.

Vision validation

The refined pipelines pass generated images through a vision-language model to verify the output matches your request before saving.

Dashboard

Upload, import, generate, and manage

Upload

Drag-and-drop, paste from clipboard, or pick from your filesystem. Images are stored per-project with thumbnails auto-generated.

Import

Import images from any URL. The server downloads, stores, and generates thumbnails automatically.

Generate

Select a pipeline, configure parameters, optionally apply a template, and generate. Pipeline runs show progress in real time.

Image detail

Click any image for a full-size preview with metadata, generation info (model, seed, sampler, steps, CFG, duration), tags, and actions — copy link, download, edit, save as template, or delete. Deep-link to any image via ?id= query parameter.