Sumith Portrait Studio

Personalized text-to-image generation with Sumith Reddy's DreamBooth Stable Diffusion 1.5 checkpoint. The subject phrase is photo of sumithreddy person.

Install this repair

In the existing Sumith0102/personal_project Space, upload all five files from this ZIP into the root directory, replacing existing files:

• app.py
• prompts.py
• convert_checkpoint.py
• requirements.txt
• README.md

Keep the Space on its existing ZeroGPU hardware. Commit the upload and allow the Space to rebuild. The old template used SDXL Turbo and omitted the ZeroGPU decorator; this package uses the personalized model and declares a GPU function.

The app downloads model.ckpt from Sumith0102/PersonalisedAiAvatarGenerator at startup, converts it to safetensors using torch.load(weights_only=True), then loads the model before the GPU worker starts. The first startup downloads several GB. The model repository is currently public. If you make it private, add an HF_TOKEN secret with read access. Never embed tokens in source.

The model, its components, and the enabled safety checker must all load before the app starts. The source uses Gradio 6.5.1, Python 3.12, and PyTorch 2.8.0. No paid hardware is requested by this package. ZeroGPU usage is governed by the host's quotas.

Verification status

Source syntax and ZeroGPU wiring checks passed. This repair has not been uploaded to the Space, built there, or tested for real image generation. The current published Space remains in RUNTIME_ERROR. Test a fictional scene after rebuilding, check likeness and image download, then connect the verified app URL to the portfolio. Do not overwrite the existing Mood Atlas project.

Optional configuration: MODEL_REPO_ID, MODEL_FILENAME, BASE_MODEL_ID, MODEL_PATH, HF_TOKEN. Defaults select the user's uploaded checkpoint and Stable Diffusion 1.5 components.

Images are synthetic. Prompts and results are processed by the hosting service. The app serializes generation, queues up to four requests, and periodically deletes temporary image outputs. It is a portfolio demonstration rather than a high-traffic service.

Reference: https://huggingface.co/docs/hub/spaces-zerogpu