Don't Pay for AI! 5 Open Source Alternatives That Are Better

Summary

The video presents five open-source projects as lower-cost alternatives to paid AI subscriptions. Locally run tools can keep data on the user’s computer, though free local alternatives may not match the quality or polish of paid services.

1. FreeLLMAPI — Coding assistants such as Claude Code or Codex

FreeLLMAPI helps users manage access to more than 34 free large-language-model providers. The presenter says these providers collectively offer up to 7 billion tokens per month and include models such as Gemini, GPT, DeepSeek, Qwen, and GLM.

A separate, more detailed setup guide is referenced for finding providers and connecting them to coding tools. Unlike the other alternatives shown, FreeLLMAPI relies on external providers, so it does not offer fully local processing.

2. FluidVoice — Wispr Flow and Granola-style transcription

FluidVoice provides speech-to-text and meeting-note features. The demonstration covers choosing and downloading a transcription model, selecting a language, granting microphone access, dictating through a live recording interface, and saving meeting notes.

The presenter says FluidVoice can also summarize notes using a local model and work offline. The model used in the demonstration did not handle Chinese well, so users may need a different model for multilingual transcription.

3. VoiceStudio — ElevenLabs-style voice cloning

VoiceStudio is shown cloning a voice from about 75 seconds of audio and generating speech from a supplied script. The presenter also mentions voice dubbing and dictation.

The local result works but sounds less natural than ElevenLabs. Its main advantages are local processing, privacy, and avoiding subscription credits.

4. ACE-Step 1.5 — Suno-style music generation

ACE-Step 1.5 is presented as a local music-generation model that can run on Mac and other supported devices. The presenter tests it with the same lyrics and style prompts used for a Suno example.

He judges the generated song’s quality as weak in this brief test, but suggests the tool may be useful for instrumental or background music. The result followed roughly 20–30 minutes of setup or experimentation.

5. InvokeAI — Image generation

InvokeAI is demonstrated generating images locally as an alternative to paid image-generation tools. The sample thumbnail is described as usable overall, but it has visible flaws, including missing or poorly rendered hands, unclear faces, and cut-off text.

The presenter suggests that trying another model or version might improve the results.

Additional Demonstrations

Make sponsorship

The presenter builds an approval workflow using Make’s AI assistant. The workflow searches a customer record, proposes a plan change, requests human approval through Telegram, updates the record only after approval, and logs the change.

The example is presented as a portfolio project demonstrating AI-assisted workflows with human oversight.

Community promotion

The presenter also promotes a community focused on AI development, automation, agents, and building AI products.

Main Speaker and Sources

• Main speaker: Eric, from the Eric Tech channel.
• Other sources and tools mentioned or demonstrated: FreeLLMAPI, FluidVoice, VoiceStudio, ACE-Step 1.5, InvokeAI, and the sponsored platform Make.