Chewy

A terminal UI for AI image and video generation with Stable Diffusion, FLUX, Wan, DALL-E, Imagen, and more.

Built with the Charm Ruby ecosystem (bubbletea, lipgloss, bubbles). Supports local generation via stable-diffusion.cpp and 5 cloud providers.

Website • Documentation • Issues

---

Install

brew install Holy-Coders/chewy/chewy

This installs both chewy and sd (the stable-diffusion.cpp inference engine with Metal GPU acceleration).

Manual install

Requirements: Ruby 4.0+, stable-diffusion.cpp

git clone https://github.com/Holy-Coders/chewy.git
cd chewy
bundle install
ruby chewy.rb

Set SD_BIN to point to your sd binary if it's not on your PATH.

Usage

chewy              # Launch the TUI
chewy list          # List all generated images
chewy delete FILE   # Delete a generated image
chewy help          # Show help

Providers

Chewy supports 6 image generation backends. Press ^y to switch providers.

Provider	Models	Type
Local (sd.cpp)	SD 1.x/2.x/3.5, SDXL, FLUX, Wan Video (.gguf/.safetensors/.ckpt)	Local
OpenAI	GPT Image 1, DALL-E 3, DALL-E 2	API
Gemini	Imagen 3, Imagen 3 Fast, Gemini 2.0 Flash	API
HuggingFace	FLUX.1 Schnell/Dev, SDXL, SD 3.5 Large, HiDream	API
OpenAI-Compatible	Any model via custom endpoint	API

API keys are entered in-app (stored securely with chmod 600) or via environment variables (OPENAI_API_KEY, GEMINI_API_KEY, HUGGINGFACE_API_KEY).

Keyboard shortcuts

Key	Action
tab	Cycle focus between prompt, negative prompt, and params
enter	Generate image (when in prompt/negative)
up/down	Cycle through prompt history (in prompt field)
^y	Switch provider
^n	Open model picker
^d	Download models (HuggingFace / CivitAI)
^t	Theme picker (10 built-in + custom themes)
^a	Gallery
^g	Generation history
^b	Browse for init image (img2img)
^v	Paste image from clipboard (img2img)
^u	Clear init image
^l	LoRA selection
^p	Presets
^e	Open last image in viewer
^f	Fullscreen image preview
^x	Cancel generation
^w	Clear prompt and image (start fresh)
^r	Randomize seed
alt+h	Search prompt history
alt+q	Queue current generation
alt+s	Seed sweep (4 random seeds)
alt+e	AI enhance prompt
alt+n	AI generate negative prompt
alt+r	AI random creative prompt
F1	Help overlay (all shortcuts)
^q	Quit

In the gallery: c copies the selected image to your clipboard, u upscales it (Real-ESRGAN 4x), p loads its prompt and settings.

Models

Place .gguf, .safetensors, or .ckpt model files in ~/.config/chewy/models (or set CHEWY_MODELS_DIR). Full filenames are shown in the model picker.

Chewy also scans for models from:

• DiffusionBee (~/.diffusionbee)
• Draw Things (~/Library/Containers/com.liuliu.draw-things/Data/Documents/Models)

Press ^d inside chewy to browse recommended starter models or search HuggingFace and CivitAI. Curated picks include SD 1.5, SD 3.5 Medium, SDXL Turbo, DreamShaper, and FLUX.1 Schnell.

FLUX models

FLUX models require companion files (clip_l, t5xxl, vae). Chewy will automatically download these when you first try to generate with a FLUX model. You'll need a HuggingFace token with read access.

Chewy also detects and runs Chroma (uncensored Apache-2.0 FLUX derivative), FLUX.1 Kontext (reference-image edits), FLUX.2 Klein/Dev, Z-Image, and Qwen-Image — companion files are downloaded automatically per model family.

Upscaling

In the gallery, press u on any image to upscale it 4x with Real-ESRGAN. The upscaler model is downloaded automatically on first use. Output is saved alongside the original.

Generation Queue & Seed Sweep

• alt+q — queue the current prompt for generation. When the current one finishes, the next is launched automatically. Keep stacking prompts while one is running.
• alt+s — seed sweep: generate 4 variants of the current prompt with random seeds, queued back-to-back.

Prompt Search

Press alt+h in the prompt field to open a fuzzy-searchable view of your prompt history. Type to filter, up/down to select, enter to load.

Video Generation (Experimental)

Chewy supports video generation via Wan 2.1 models using sd.cpp's -M vid_gen mode. Select a Wan model and Chewy will show a VIDEO badge, add Frames/FPS parameters, and output frame sequences with an in-terminal video player.

Note: Video generation is experimental and works best with dedicated GPU hardware. The Wan 1.3B model requires ~12GB RAM just for weights, plus additional memory for video frame generation. CPU-only generation is very slow (minutes per frame). Companion files (t5xxl, clip_vision, vae) are auto-downloaded on first use.

Video presets are available under ^p: Quick Preview, Standard, High Quality, and Img2Vid.

LoRAs

Place LoRA .safetensors files in ~/.config/chewy/loras (or set CHEWY_LORA_DIR). Press ^l to open the LoRA selector where you can toggle LoRAs on/off, adjust weights (0.0-2.0), and download new ones.

Press d in the LoRA panel to browse recommended LoRAs or search HuggingFace and CivitAI. Recommended picks include Detail Tweaker, LCM LoRA (for fast generation), Pixel Art, and Papercut styles. Chewy warns if a LoRA doesn't match your model's architecture (e.g. SDXL LoRA on SD 1.5 model).

Inpainting

Inpainting lets you selectively regenerate parts of an image while preserving others (e.g. change the background but keep the face). Set an init image (^b), then in the params panel navigate to Mask and:

• Press g to auto-generate a center-preserve mask (keeps the face, regenerates the rest)
• Press p to open the mask painter — click to paint which areas to regenerate
• Press enter to browse for a custom mask image

White pixels in the mask = regenerate, black = keep. The "Inpaint - Face Preserve" preset configures everything automatically.

Themes

10 built-in color themes: Midnight (default), Dracula, Catppuccin, Tokyo Night, Gruvbox, Nord, Rose Pine, Solarized, Horizon, and Light. Press ^t to switch.

Custom themes: Drop a YAML file in ~/.config/chewy/themes/ with 13 color keys and it appears in the theme picker. Example ~/.config/chewy/themes/cyberpunk.yml:

primary: "#00FF41"
secondary: "#008F11"
accent: "#FF0090"
success: "#00FF41"
warning: "#FFD300"
error: "#FF0000"
text: "#E0E0E0"
text_dim: "#808080"
text_muted: "#505050"
surface: "#0D0D0D"
border_dim: "#333333"
border_focus: "#00FF41"
bar_text: "#000000"

CLI commands (help, list, etc.) also use your selected theme colors. An animated pixel-art splash screen greets you on startup.

Presets

Press ^p to open the preset picker. Built-in presets include fast, quality, portrait, flux-fast, and flux-quality configurations. Save your own custom presets with your preferred settings.

AI Prompt Enhancement

Chewy can enhance your prompts, generate negative prompts, and create random creative prompts using a local or cloud LLM.

• alt+e — Enhance your prompt (adds detail, lighting, composition, quality tags)
• alt+n — Auto-generate a negative prompt from your positive prompt
• alt+r — Generate a random creative prompt from scratch
• Click the ✨ Enhance / ✨ Generate button in the prompt box

LLM priority: Local (ollama/LM Studio) → OpenAI → Gemini → rule-based fallback. For the best local experience: brew install ollama && ollama pull llama3.2:3b.

Prompt History

Chewy remembers your prompt history across sessions. Press up/down in the prompt field to cycle through previous prompts. History is loaded from your generation metadata on disk, so past prompts are available even in a fresh session.

Samplers & Schedulers

14 samplers (euler, euler_a, heun, dpm2, dpm++2s_a, dpm++2m, dpm++2mv2, ipndm, ipndm_v, lcm, ddim_trailing, tcd, res_multistep, res_2s) and 9 schedulers (discrete, karras, exponential, ays, gits, sgm_uniform, simple, smoothstep, kl_optimal). Configurable thread count for local generation.

img2img

Press ^b to browse for an input image, or ^v to paste from clipboard. Adjust the strength parameter (0.0-1.0) to control how much the output differs from the input. Use ^u to clear and go back to txt2img.

Configuration

Config is stored at ~/.config/chewy/config.yml. Presets are stored at ~/.config/chewy/presets.yml.

Env var	Description
SD_BIN	Path to the sd binary
CHEWY_MODELS_DIR	Model directory (default: ~/.config/chewy/models)
CHEWY_OUTPUT_DIR	Output directory (default: ./outputs)
CHEWY_LORA_DIR	LoRA directory (default: ~/.config/chewy/loras)

Contributing

Setup

git clone https://github.com/Holy-Coders/chewy.git
cd chewy
bundle install

You'll need stable-diffusion.cpp built and on your PATH (or set SD_BIN):

git clone --recursive https://github.com/leejet/stable-diffusion.cpp
cd stable-diffusion.cpp
mkdir build && cd build
cmake .. -DSD_METAL=ON    # macOS with Metal GPU
cmake --build . --config Release
cp bin/sd-cli /usr/local/bin/sd

Project structure

• chewy.rb — entrypoint (CLI routing + app launch)
• lib/chewy.rb — main Chewy class (init, update, view)
• lib/chewy/*.rb — 16 modules mixed into the Chewy class
• Gemfile / Gemfile.lock — Ruby dependencies
• VERSION — semver, read at runtime and used by CI for releases
• Formula/ — Homebrew formulas (sd-cpp and chewy)
• docs/ — Landing page and documentation (GitHub Pages)
• logo.jpeg — the logo

Dependencies

Gem	Purpose
bubbletea	Elm Architecture TUI framework
lipgloss	Terminal styling and layout
bubbles	TUI components (list, text input, spinner, progress)
chunky_png	PNG reading for terminal image rendering

Making changes

1. Fork the repo
2. Create a branch (git checkout -b my-feature)
3. Make your changes to chewy.rb
4. Test locally with ruby chewy.rb
5. Open a PR

Releasing

Bump the version in VERSION and push to main. CI will automatically:

1. Create a GitHub release
2. Update the Homebrew formula with the new sha256

License

MIT - Copyright (c) 2026 Holy Coders