New on Pinokio
Automatically discovered via GitHub
Automatically discovered via GitHub
crispz does one thing: it enlarges an image and puts real detail back into it, 100% local, no ComfyUI, no SwarmUI, no cloud. Two stages: Real-ESRGAN (through spandrel) for the true pixel enlargement, with overlap-add tiling and linear feathering: no visible seams, whole images up to 2048 px on a 5090, tiled above. Z-Image Turbo img2img at a low denoise, the pass that reinjects texture, skin, fabric and foliage without touching the composition. You get a Gradio UI, a scriptable CLI and an interactive CLI with saved preferences, for one image or a whole folder. It is the upscaler that every cris
comics2crispz turns a script into a comic book: chapters, a flatplan of facing pages, panels drawn by a local image model, balloons lettered in place, a print-ready PDF. It draws nothing itself: it drives the crispz studios already on your machine (crispz-studio, crispz-klein, crispz-flux3, crispz-qwen-edit, Fooocus2026...) through a small JSON protocol, so the GPU is shared and the model stays warm. Pure Python, runs on CPU. Write it as text. A .czs script (=== chapter ===, --- page 4-grid, [pn1] what to draw, Lea: what she says) or a single .md bundle with the settings, the Visual Bible and
crispz-ideogram is a Fooocus-style web studio for Ideogram 4.0, the open-weights model that renders text in images best: posters, logos, packaging, slides, covers, dense typography. Everything runs on your own GPU, nothing leaves the machine. Read the licence first. Ideogram 4.0's weights are under the Ideogram 4 Non-Commercial licence: personal use and research only; any commercial use of the self-hosted weights needs a licence negotiated with Ideogram. For commercial design work, the same workbench exists with an Apache 2.0 model: crispz-moss. JSON captions, the way the model was trained. An
crispz-flux3 is a Fooocus-style web studio for FLUX.3 Image Klein, the 5.6B image model Black Forest Labs released on October 15. Everything runs on your own GPU: text to image, multi-reference editing, img2img and upscale, inpaint and outpaint, 2K and 4K native renders, LoRAs, single-file checkpoints from CivitAI (bf16, INT8, GGUF), a CLI and a JSON protocol for automation. Pinokio launcher: Source: Install (Pinokio) In Pinokio, open Discover and paste Click Install. The launcher clones the app, creates a Python 3.11 venv, installs diffusers 0.42.1 (the first release with FLUX.3), PyTorch 2.1
crispz-qwen-edit is a Fooocus-style web studio for Qwen-Image (text to image, with the best in-image text of the 20B generation) and Qwen-Image-Edit-2509, the model that edits a picture from a plain instruction: "add heavy rain, keep everything else", "she looks to the left", "make it night". Both are Apache 2.0. Everything runs on your own GPU, no ComfyUI, no cloud. Edit tab: an image plus an instruction in plain words; the result keeps what you did not ask to change, and the previous version stays one click away. Text to Image with styles, wildcards, tag autocomplete, prompt variants, live p
crispz-krea2 is a Fooocus-style web studio for Krea 2, the open-weights successor of FLUX.1 Krea, wrapped in the same workbench as crispz-studio. Everything runs on your own GPU, no ComfyUI, no cloud. Text to Image with styles, wildcards, tag autocomplete, prompt variants, live preview and Stop; Upscale after generate chains each image through ESRGAN. Models: single-file checkpoints from CivitAI, LoRAs with hot swap, a CivitAI search with previews and trigger words. Job queue and X/Y/Z grids for overnight batches and side-by-side comparisons. Prompt AI, local: Describe, Improve and Vision Mix
crispz-krea is a Fooocus-style web studio for FLUX.1 Krea [dev], the Krea and Black Forest Labs model trained against the "AI look": natural skin, real light, no plastic sheen. Same workbench as crispz-studio, with the Flux pipeline inside. Everything runs on your own GPU, no ComfyUI, no cloud. Text to Image with styles, wildcards, tag autocomplete, prompt variants, live preview and Stop. Upscale and enhance: Real-ESRGAN plus a Flux refine pass, tiled up to 4K, one-click Vary, Reframe / outpaint that generates the missing bands instead of cropping. Models: single-file Flux checkpoints from Civ
crispz-klein is a Fooocus-style web studio for FLUX.2 Klein 4B, Black Forest Labs' Apache 2.0 model distilled to 4 steps. One model covers the whole surface: text to image, multi-reference editing and inpaint run in the same pipeline, with no second model loaded. Measured on an RTX 5090: 14.9 GB of VRAM for everything, 1024x1024 in 2.0 s, an edit with one reference in 3 s. Generate Edit (1 reference) Inpaint The middle image is the point: "make it daytime, sunny blue sky", and the character stays pixel-identical, same face, same pose, same clothes. Edit with up to 4 references in the tab (more
crispz-studio is a Fooocus-style web studio for Z-Image Turbo, the distilled 6B model that renders a clean 1024 image in about 8 steps at guidance 0. It is the root of the crispz family: the other studios (klein, krea, qwen-edit, flux3, moss, ideogram) are forks of this one, so what you learn here works everywhere. Everything runs on your own GPU, no ComfyUI, no cloud. Text to Image with styles (277, searchable, with previews), wildcards, tag autocomplete, prompt variants, a live preview and Stop. Upscale and enhance: Real-ESRGAN plus a Z-Image refine pass, tiled up to 4K, with one-click Vary
crispz-moss is a Fooocus-style web studio for MOSS-Image, the 5B open-weights model OpenMOSS (Fudan) released in October under Apache 2.0. It is the text-in-image and design studio of the crispz family: posters, logos, packaging, slides, covers, dense typography. Everything runs on your own GPU, nothing leaves the machine. What makes it different from a prompt box: Layouts, not luck. You draw the text boxes on a canvas, type the exact words, and the model renders those words and nothing else. Nine design presets at their native sizes (A-series poster, logo, business card, book cover, 16:9 slid
Automatically discovered via GitHub
Automatically discovered via GitHub
Automatically discovered via GitHub
Batch image captioning with a vision model served by Ollama. One core, four front ends: a command line, a Tkinter desktop app (standard library only), a NiceGUI web UI (--ui web), and a Gradio app made for Hugging Face Spaces. The Ollama layer reuses proven patterns from crispz-studio (cz_ollama.py): vision detection through /api/show with a name-based fallback, JPEG downscaling before upload, stripping of blocks from "thinking" models, keep_alive / CPU mode so the model does not hog VRAM. Features Connect to any Ollama server (configurable URL, local or remote) Automatic list of installed mod
Small free BPM and track check !
A one-click way to install my fork of Fooocus 2.5.5, with the quality-of-life additions I kept missing in upstream: 💾 Save Preset directly from the UI (no JSON editing) 🏷️ LoRA triggers auto-pulled from local .safetensors metadata + CivitAI 🧩 Embeddings panel: 5 slots, prompt or negative injection 🎲 Wildcards editor (browse, edit, create token files in-app) 🎨 CivitAI Model Settings — checkpoint triggers, consensus settings, save-as-preset 📐 Custom Resolution + Vary-with-aspect-ratio override (any ratio, snapped to /64) 🖼️ Asset Browser (PhotoSwipe gallery for outputs + model previews) �