Flowith
Flowith is an agentic AI workspace that combines a whiteboard-style canvas, access to 40+ image and video models, and mini apps for design, e-commerce, and social content.

The long-standing browser UI for running Stable Diffusion locally
AUTOMATIC1111's Stable Diffusion Web UI is a browser-based front end for Stable Diffusion, built with the Gradio library and distributed through GitHub. It is aimed at people who want to run image generation on their own hardware and control the process in detail, rather than pay for a hosted service.
The project centers on the two original Stable Diffusion modes: txt2img, which generates an image from a text prompt, and img2img, which transforms an existing image. Around those core modes it adds inpainting, outpainting, and a color sketch tool, so the same interface handles both fresh generation and edits to existing artwork. A negative prompt field lets users list what they do not want in the output, and a Styles feature saves parts of a prompt for reuse through a dropdown.
Prompt handling goes well beyond a single text box. Attention syntax lets users weight parts of a prompt, written as ((tuxedo)) or (tuxedo:1.21), and selected text can be adjusted with Ctrl+Up and Ctrl+Down, or Command+Up and Command+Down on macOS. Prompts are not limited to the 75 tokens of the original Stable Diffusion. Composable-Diffusion support allows multiple prompts in one generation, separated by an uppercase AND, with per-prompt weights. Generation parameters are saved with each image, in PNG chunks for PNG files and EXIF for JPEG, and dragging an image into the PNG info tab restores those parameters into the UI. That makes it possible to revisit an earlier result and reproduce or adjust it.
The Extras tab bundles several neural networks for improving output. GFPGAN and CodeFormer handle face restoration, while RealESRGAN, ESRGAN, SwinIR, Swin2SR, and LDSR provide upscaling through different approaches. ESRGAN in particular works with a large number of third-party models. For systematic experimentation, the X/Y/Z plot draws a three-dimensional grid of images generated with different parameters, and Loopback runs img2img processing repeatedly. Highres Fix produces higher-resolution images in one click, and a checkpoint merger tab can combine up to three checkpoints into one.
Textual Inversion support lets users create as many embeddings as they want, name them freely, and combine multiple embeddings with different numbers of vectors per token. The project reports that embeddings can be trained on 8GB of VRAM, with some reports of 6GB working. Hypernetworks and Loras are also supported, and a separate UI lets users choose embeddings, hypernetworks, or Loras with previews before adding them to a prompt. A different VAE can be loaded from the settings screen, and checkpoints can be reloaded on the fly.
The interface supports 4GB video cards, with reports of 2GB cards working, and processing can be interrupted at any time. A progress bar shows an estimated completion time and a live preview of the image being generated; a separate neural network can produce those previews with almost no VRAM or compute cost. Sampling method selection includes adjustable sampler eta values and more advanced noise settings. Adding the --xformers command-line argument can produce a major speed increase on select cards.
The project supports custom scripts and a large collection of community extensions. One extension adds a history tab for viewing, directing, and deleting images inside the UI; another, Aesthetic Gradients, generates images with a specific aesthetic using CLIP image embeddings. An API is available, and the interface includes an option to generate forever. DeepDanbooru integration creates danbooru-style tags for anime prompts, and a CLIP interrogator button attempts to guess a prompt from an image.
Installation is handled by a one-click script, but Python and Git still need to be installed first. Running arbitrary Python code from the UI requires launching with the --allow-code flag, which is worth understanding before enabling it. Many of the more specialized capabilities come through third-party models and extensions rather than the core install, so the experience depends partly on what users add. The project is distributed on GitHub, and its feature set has grown through contributions from a long list of developers and community members.
Subscription pricing is not listed.
Features, pricing and availability may change. Always verify details on the official website. Maintained by steven · Last updated Oct 11, 2026 →
Flowith is an agentic AI workspace that combines a whiteboard-style canvas, access to 40+ image and video models, and mini apps for design, e-commerce, and social content.
Fine-tune open models faster with far less memory
Groq is an AI inference cloud for developers and enterprises that need low-latency model responses at scale, delivered through custom LPU hardware and a tiered platform spanning bare metal, inference, and enterprise controls.
Jan is an open-source desktop chat application that lets users run open models on their own hardware or connect to hosted providers like OpenAI, Anthropic, and Google. It targets privacy-conscious individuals and teams who want a friendlier local LLM experience than command-line tooling offers.