Andrew Warner of The Next New Thing tours ten open-source projects that can replace parts of paid AI workflows. Running software yourself can mean more control and privacy, but it does not automatically provide free models, equivalent quality, or commercial-use rights.

Ten alternatives

  1. VoiceStudio (00:27) — local voice cloning, narration, and dubbing instead of ElevenLabs; check the chosen voice model’s license before commercial use.
  2. Odysseus (02:42) — private chat, research, memory, and agents instead of some ChatGPT workflows; the local model and your hardware determine capability.
  3. FreeLLMAPI (05:51) — routes requests across providers’ free model allowances; not a free Claude subscription, and provider limits can change.
  4. Vane (09:27) — self-hosted, cited web research in the Perplexity mold; local models need adequate hardware, while cloud keys may cost money.
  5. InvokeAI (11:51) — an image-generation and editing studio with a layered canvas and repeatable workflows; fast local use calls for a capable GPU.
  6. ACE-Step (13:57) — generate songs and vocals locally instead of Suno; Warner says the output needs more work and may not match Suno’s quality.
  7. OpenShorts (19:30) — cut long videos into captioned vertical clips; self-hosting avoids the hosted trial’s limits but takes compute and manual reframing.
  8. Dyad (23:51) — build apps from plain-language prompts on your own machine instead of paying a builder’s credits; the underlying model can still cost money.
  9. OpenManus (26:15) — plan and execute browser, research, coding, and file tasks with a model you supply; API usage is not inherently free.
  10. Upscayl (29:06) — batch-upscale photos locally rather than subscribing to a hosted image enhancer; a compatible GPU matters, and cloud face repair is separate.

The useful question is not “Which subscription can I cancel?” but “What quality, compute, licensing, and setup costs am I taking on in exchange for control?”