Prompt & Done · FREEMAC

4 AI subscriptions, 4 free tools that run on your Mac

Every tool here is one we ran ourselves on an Apple-silicon Mac, with the real numbers from our tests. All four run locally: no account, no upload, no monthly bill. You need a Mac with an M-series chip, some disk space for the model weights, and a terminal.

Prices and licenses checked 7 Nov 2026 on the official pages. Speeds depend on your Mac; ours were an M5 Max (64 GB) and an M4 Pro (24 GB).

1 Transcription → mlx-whisper (instead of e.g. Otter Pro, $16.99/mo)

OpenAI's Whisper large-v3-turbo running on Apple's MLX. Our test: a 60 min 33 s LibriVox audiobook (Alice in Wonderland, ch. 1-4) transcribed in 28.2 s on an M5 Max, 4.2% word error rate against the book text. The first run downloads the model weights.

brew install ffmpeg
pip install mlx-whisper
mlx_whisper interview.mp3 --model mlx-community/whisper-large-v3-turbo --language en --output-format txt

mlx-whisper 0.4.3 (MIT) · pypi.org/project/mlx-whisper · model mlx-community/whisper-large-v3-turbo (Whisper weights: MIT). Accuracy drops with noise, accents and crosstalk.

2 Scanning / OCR → indra-ocr (instead of a scanner-app plan, e.g. Genius Scan Ultra, $39.99/yr)

Point it at a phone photo or a scanned PDF; it returns clean Markdown with headings and tables. Our test: a crooked, shadowed photo of an order page came back word-for-word with the full price table in 0.5 s (73 ms of OCR) on an M5 Max; upside-down and sideways photos gave identical output. The model is about 76 MB.

pip install indra-ocr
indra-ocr receipt-photo.jpg --timings

indra-ocr 1.1.2 · github.com/datalab-to/indra · code Apache 2.0; model weights under a modified OpenRAIL-M license (free for personal use and small companies, see the repo). Handwriting is weak; we didn't claim it. Our test photo was a simulated phone shot.

3 AI images and photo edits → FLUX.2 [klein] 4B via mflux (instead of e.g. Midjourney Basic, $10/mo)

Black Forest Labs' small FLUX.2 model, 4 steps, fully local. Our test made the bakery photo from one sentence (46 s) and turned it into a snowy night shot with a second sentence (79 s) on an M4 Pro with 24 GB. First run downloads about 15 GB of weights and peaks around 20 GB of memory, so 24 GB+ is comfortable.

uv tool install --upgrade mflux
mflux-generate-flux2 --model flux2-klein-4b --steps 4 --width 864 --height 1536 --seed 6 --prompt "A bakery storefront seen straight on ... photograph" --output base.png
mflux-generate-flux2-edit --model flux2-klein-4b --steps 4 --image-paths base.png --seed 1 --prompt "The same bakery storefront straight on, at night in heavy snowfall ... photograph" --output snow.png

FLUX.2-klein-4B: Apache 2.0 (model card; the 9B variant is non-commercial) · mflux 0.24.0 (MIT) · github.com/mflux-community/mflux. On mflux 0.23.0 we needed uv tool install --with "pillow<12" mflux.

4 Voice cloning → Chatterbox (instead of e.g. ElevenLabs Starter, $6/mo)

The voice in the reel is Chatterbox: we generated the whole narration on our Mac from a 14-second reference clip of the channel's narrator, 4 takes per sentence, then kept the takes that passed a word-for-word speech-recognition check. Earlier test (Chatterbox Turbo, M4 Pro): 5 s of speech in 3.4 s. Only clone voices you have the right to use; Chatterbox stamps its built-in Perth watermark on every clip.

python3.11 -m venv .venv && source .venv/bin/activate
pip install chatterbox-tts
python - <<'PY'
import torchaudio as ta
from chatterbox.tts import ChatterboxTTS
model = ChatterboxTTS.from_pretrained(device="mps")
wav = model.generate("Every word you heard was free.", audio_prompt_path="my_voice_14s.wav")
ta.save("out.wav", wav, model.sr)
PY

chatterbox-tts 0.1.7 (MIT) · huggingface.co/ResembleAI/chatterbox · Python 3.11, about 3 GB of weights.

5 The math

Otter Pro $16.99 + Genius Scan Ultra $39.99/yr (≈ $3.33/mo) + Midjourney Basic $10 + ElevenLabs Starter $6 = $36.32 a month at monthly-billing list prices (7 Nov 2026). Annual plans and promos are cheaper; the paid apps also do things these tools don't (sync, sharing, editing UIs, cloud speed on slower Macs).

otter.ai/pricing · Genius Scan pricing · Midjourney plans · elevenlabs.io/pricing