Use the code, then check it against the evidence.

Choose a project

1. 01

   Remotion templates

   Agent-friendly video templates that run locally through Codex or Claude Code.

   Open repositorySee how the templates work
2. 02

   Instavar MCP server

   Free Remotion template tools for local agents, with optional hosted Studio workflows.

   Open repositorySee the MCP workflow
3. 03

   Instavar Voice Evaluation

   Shared contracts and utilities for corpus checks, lifecycle runs, objective metrics, blind listening, resume, packaging and evidence.

   Open repositoryInspect the evaluation framework
4. 04

   Cartesia Sonic evaluation

   Evidence-first Sonic 3.6 streaming and voice-clone experiments, with provider-managed fine-tuning kept distinct from unsupported local LoRA or full SFT.

   Open repositoryReview the TTS comparison
5. 05

   Breeze TTS 2 fine-tuning toolkit

   Source-only LoRA and full-SFT training, checkpoint lifecycle and matched evaluation tools for Breeze TTS 2. Model weights remain under BreezeBlue's separate license.

   Open repositoryRead the Breeze experimentOpen the LoRA modelOpen the full-SFT model
6. 06

   Qwen3-TTS LoRA fine-tuning

   Tools, evaluation steps and operating notes for adapting Qwen3-TTS voices.

   Open repositoryRead the fine-tuning findings
7. 07

   IndexTTS2 fine-tuning

   A public training pipeline with the failure modes found during real fine-tuning runs.

   Open repositoryRead the implementation findings
8. 08

   F5-TTS LoRA fine-tuning

   PEFT adapter training and inference for F5-TTS, validated on a single-GPU workflow.

   Open repositoryRead the training findings
9. 09

   CosyVoice3 LoRA fine-tuning

   PEFT-based LoRA integration, merged-checkpoint probes and retained negative-result evidence from our CosyVoice3 runs.

   Open repositoryRead the CosyVoice3 findings
10. 10

    Audio8 TTS LoRA fine-tuning

    LoRA training, reload and evaluation workflows for Audio8 TTS on CUDA and Apple Silicon MPS.

    Open repositoryRead the Audio8 findings
11. 11

    MicroZoom INT4

    An experimental INT4 inference and LoRA training port tested on a 24 GB RTX 3090 Ti.

    Open repositoryRead the deployment findings