I'm vibe coding an image + video + audio generation app with comfyui backend
retained text
I always loved how nice to look at was automatic1111 stable diffusion webui and how easy it was for a beginner to just press generate and watch your image being made. But the need for more user control that came with comfyui basically replaced the webui space where it's no longer being used as the need for dynamic control over workflows keep changing at a rapid pace and more models are being released every day. So I been vibe coding this app for weeks now slowly sprinkling my ideas in it and fixing what I think doesn't work , I'm basically just the creative lead on this thing. It works as a frontend wrapper for comfyui where you just need to press a button and not worry about the workflow behind it. So here is the repo make with it what you want I hope you help me improve it or just fork it and make something new by yourself: [https://github.com/itsbalanse/try2create](https://github.com/itsbalanse/try2create) So far I made the following features: **Gallery / prompt reader** * Open any folder and see the prompt, settings, LoRAs and tags inside each picture: A1111, ComfyUI, NovelAI, InvokeAI and others. * Search by words or by meaning, filters, ratings, duplicate hiding, and a blur toggle. **Generate** * One prompt box. It picks the model for you: Krea 2 for people and most pictures, Qwen Image for creatures and extreme looks, MOSS when you need exact text on a poster or cover. * Or run your own ComfyUI workflow. * LoRAs are listed per model, so you only see the ones that work with it. **Edits** * **One-click apps:** about 25 recipes like relight, new camera angle, plushie, product shot, headshot, outfit swap and day to night. Pick a picture, press the button. * **Change anything:** type “make it night” or “remove the car” and Qwen Image Edit does it. * **Designs & exact text:** posters and thumbnails with the words you actually typed. You can also box one part of any picture and redo just that. * **Text over a picture:** memes, captions and speech bubbles as real layers (exports a .psd). * **Upscale:** bigger, with redrawn detail and fixed faces. **Video with sound (MiniMax H3)** * **Animate a picture:** pick one of 18 idea cards (viral dance, kung fu, bullet time, rain dance…) and a camera move. A local LLM writes the video prompt and H3 makes the video and its sound. * **Long videos up to a minute:** planned as one story, made in parts and joined. * **Star in a video:** click the person in a clip and swap in your character. * **Make a story:** short TikTok-style drama videos from one idea. * **Recreate:** load any video’s settings and prompt and make it again with another model or quality. * **Make it crisp:** redraws a video at 2K with real new detail (NVIDIA’s Open-H3-Regenerate), keeping the face, motion and sound. **Sound** * **Music:** songs with lyrics and an AI-made cover. * **Drama:** audio scenes with several voices, written for you from one line. **Characters & story** * Create characters or invent a whole cast from an idea, with backstories, character sheets and a 360° turntable. * **Vision board:** lay out cards and turn them into scenes. * **Clef mode:** a scenario that keeps going picture after picture. * **Pixel art:** made into real, clean pixels, not just a pixel-art style. **Small stuff that makes it feel nice** * Ctrl+K command palette. * A job tracker at the bottom showing what’s generating, which step it’s on and the time left. * Pop-up cards when something finishes. * Prompt writing and image descriptions run locally too, no API keys. I’ve only tested it on my own PC: Windows 11, RTX 3090 (24 GB), 128 GB RAM, with ComfyUI Desktop. The picture stuff should run on less. The video stuff is heavy.
document fetched 2026-10-29 00:33 · canonical https://reddit.com/r/StableDiffusion/comments/1xhr1ak/im_vibe_coding_an_image_video_audio_generation