Skip to content

The truly free, open-source, local-first, cross-platform CapCut replacement. Scriptable through a JSON-RPC API & MCP.

11.5 KB
Raw Blame
# The models Concat downloads, and where they come from.
#
# Concat ships no weights inside its bundles - a cutout network, a Kokoro
# voice bank and a whisper model together outweigh the editor many times
# over, and most people never ask for all three. They are fetched the first
# time a feature needs one.
#
# What they are fetched *from* is this file. Every model is mirrored onto a
# release of this repository, under the `release` tag below, and that mirror
# is what the app asks for; `upstream` is where the mirror was filled from
# and what the app falls back to if our own release cannot be reached. The
# point of the mirror is that a model Concat offers stays reachable, stays
# the bytes we tested, and stays under one roof, rather than depending on a
# third party's bucket keeping a path alive.
#
#   scripts/models.py --check          # this table against the engine's
#   scripts/models.py mirror --out dir # fill a mirror from upstream
#   scripts/models.py release-manifest # the manifest.json a release carries
#
# Refreshing the mirror is .github/workflows/models.yml: it fetches every
# `upstream`, checks each against `sha256`, records the digest when this
# table has none yet, and uploads the lot to the `release` tag. Bump that
# tag only when a model's bytes change - app releases point at it by name,
# so an unchanged model is never re-uploaded.
#
# `sha256` is what the app verifies a finished download against, and a
# download with no digest to check is refused, so a model added here is
# unusable until its digest is: the workflow prints what it computed, and
# that goes here and into the engine's table beside it. Hugging Face
# upstreams are pinned to a commit rather than `main`, so the bytes the
# fallback fetches are the bytes the digest names.

# The release every model is mirrored on.
release = "models-v1"

# ── Cutout ───────────────────────────────────────────────────────────────
# src/crates/concat-vision/src/models.rs

[[model]]
id = "cutout-person"
family = "cutout"
file = "rvm-mobilenetv3.onnx"
bytes = 14975696
sha256 = "88d4531297118f595bf2fd60f6f566aec2e559393802d1f436c380f0cbbd2828"
licence = "GPL-3.0"
upstream = "https://github.com/PeterL1n/RobustVideoMatting/releases/download/v1.0.0/rvm_mobilenetv3_fp32.onnx"

[[model]]
id = "cutout-object"
family = "cutout"
file = "isnet-general-use.onnx"
bytes = 178648008
sha256 = "60920e99c45464f2ba57bee2ad08c919a52bbf852739e96947fbb4358c0d964a"
licence = "Apache-2.0"
upstream = "https://github.com/danielgatis/rembg/releases/download/v0.0.0/isnet-general-use.onnx"

[[model]]
id = "cutout-brush-encoder"
family = "cutout"
file = "slimsam-77-encoder.onnx"
bytes = 23276014
sha256 = "9f8433273a6750b587779baa0cf5508111001bf7e7acfcf585d370139fd366d0"
licence = "Apache-2.0"
upstream = "https://huggingface.co/Xenova/slimsam-77-uniform/resolve/5850ab45f587c112167512ffef949107115e26a0/onnx/vision_encoder.onnx"

[[model]]
id = "cutout-brush-decoder"
family = "cutout"
file = "slimsam-77-decoder.onnx"
bytes = 16557892
sha256 = "f4514391764fbd56e08e119060d874ecd7d52994bfb1968af159e12d4943b5bb"
licence = "Apache-2.0"
upstream = "https://huggingface.co/Xenova/slimsam-77-uniform/resolve/5850ab45f587c112167512ffef949107115e26a0/onnx/prompt_encoder_mask_decoder.onnx"

# Enhance's restoration model. In the cutout family because the family is
# the engine table a row is checked against, and this one lives in the
# vision crate's beside the cutout models.
[[model]]
id = "enhance-upscale"
family = "cutout"
file = "realesr-general-x4v3.onnx"
bytes = 4866417
sha256 = "1940a93ee08283a0a7286183186357b1688fe9fa8ede74604b424586aaddf112"
licence = "BSD-3-Clause"
upstream = "https://huggingface.co/CoderViking/realesr-general-x4v3-onnx/resolve/c6a971706797c7502945a2b4c4274fce4900d4ab/realesr-general-x4v3.onnx"

# ── Text to speech ───────────────────────────────────────────────────────
# src/crates/concat-speech/src/tts.rs. Kokoro v1.0 multi-language, the
# official sherpa-onnx conversions; the two differ only in quantisation.

[[model]]
id = "kokoro-int8-multi-lang-v1_0"
family = "tts"
file = "kokoro-int8-multi-lang-v1_0.tar.bz2"
bytes = 132353410
sha256 = "72c347230107ef8a31098ec4d468a605244beef5f1ed00db97efa13518800653"
licence = "Apache-2.0"
upstream = "https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/kokoro-int8-multi-lang-v1_0.tar.bz2"

[[model]]
id = "kokoro-multi-lang-v1_0"
family = "tts"
file = "kokoro-multi-lang-v1_0.tar.bz2"
bytes = 349886476
sha256 = "60703707bc05ccd48d60e1eb723f38b07d6e474795a0b6a588d7220400e0f7a4"
licence = "Apache-2.0"
upstream = "https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/kokoro-multi-lang-v1_0.tar.bz2"

# Kyutai's Pocket TTS, the sherpa-onnx int8 conversion: English, and any
# voice it is given a few seconds of. The weights are CC-BY-4.0 (Kyutai);
# the ONNX export is KevinAHM/pocket-tts-onnx under the same terms.
[[model]]
id = "pocket-tts-int8-2026-01-26"
family = "tts"
file = "sherpa-onnx-pocket-tts-int8-2026-01-26.tar.bz2"
bytes = 98336520
sha256 = "2f3b88823cbbb9bf0b2477ec8ae7b3fec417b3a87b6bb5f256dba66f2ad967cb"
licence = "CC-BY-4.0"
upstream = "https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/sherpa-onnx-pocket-tts-int8-2026-01-26.tar.bz2"

# ── Chatterbox ───────────────────────────────────────────────────────────
# src/crates/concat-speech/src/chatterbox.rs. Resemble AI's Chatterbox
# Turbo, the quantised ONNX export: nine files that make one bundle, each
# a row since the mirror is flat. The `file` names carry the bundle's
# name; the engine puts each back under the name its graph expects.

[[model]]
id = "chatterbox-turbo-tokenizer.json"
family = "chatterbox"
file = "chatterbox-turbo-tokenizer.json"
bytes = 3562272
sha256 = "3f04e34bea22f9144d1a19151154095bc9ce0430bf421304f5797e716288a906"
licence = "MIT"
upstream = "https://huggingface.co/ResembleAI/chatterbox-turbo-ONNX/resolve/d21799bd0354adb85e348b8a0442a8405110a2cf/tokenizer.json"

[[model]]
id = "chatterbox-turbo-embed_tokens_quantized.onnx"
family = "chatterbox"
file = "chatterbox-turbo-embed_tokens_quantized.onnx"
bytes = 2887
sha256 = "0efe1bc01c2c48a98425a74444fd9887924d887f922c2722a6ec961ebb9e1db6"
licence = "MIT"
upstream = "https://huggingface.co/ResembleAI/chatterbox-turbo-ONNX/resolve/d21799bd0354adb85e348b8a0442a8405110a2cf/onnx/embed_tokens_quantized.onnx"

[[model]]
id = "chatterbox-turbo-language_model_quantized.onnx"
family = "chatterbox"
file = "chatterbox-turbo-language_model_quantized.onnx"
bytes = 279670
sha256 = "0b40581277e30b7034331ec8c3ad47ed71d321f015b387a95221e54e2fcbfde8"
licence = "MIT"
upstream = "https://huggingface.co/ResembleAI/chatterbox-turbo-ONNX/resolve/d21799bd0354adb85e348b8a0442a8405110a2cf/onnx/language_model_quantized.onnx"

[[model]]
id = "chatterbox-turbo-speech_encoder_quantized.onnx"
family = "chatterbox"
file = "chatterbox-turbo-speech_encoder_quantized.onnx"
bytes = 1205728
sha256 = "5b6f15870a43cf97892df86fc550a0ef4763522d527cde72b2a4316f80a34de4"
licence = "MIT"
upstream = "https://huggingface.co/ResembleAI/chatterbox-turbo-ONNX/resolve/d21799bd0354adb85e348b8a0442a8405110a2cf/onnx/speech_encoder_quantized.onnx"

[[model]]
id = "chatterbox-turbo-conditional_decoder_quantized.onnx"
family = "chatterbox"
file = "chatterbox-turbo-conditional_decoder_quantized.onnx"
bytes = 2202035
sha256 = "2af3b150196d9d559cd3c91e03da80eb27a466032369dc2b57ea729cddad3ebb"
licence = "MIT"
upstream = "https://huggingface.co/ResembleAI/chatterbox-turbo-ONNX/resolve/d21799bd0354adb85e348b8a0442a8405110a2cf/onnx/conditional_decoder_quantized.onnx"

[[model]]
id = "chatterbox-turbo-embed_tokens_quantized.onnx_data"
family = "chatterbox"
file = "chatterbox-turbo-embed_tokens_quantized.onnx_data"
bytes = 67297376
sha256 = "9025d04c124899823124b1d7bb7069b1f535fb8a6c2d88f97520eb6fecced986"
licence = "MIT"
upstream = "https://huggingface.co/ResembleAI/chatterbox-turbo-ONNX/resolve/d21799bd0354adb85e348b8a0442a8405110a2cf/onnx/embed_tokens_quantized.onnx_data"

[[model]]
id = "chatterbox-turbo-conditional_decoder_quantized.onnx_data"
family = "chatterbox"
file = "chatterbox-turbo-conditional_decoder_quantized.onnx_data"
bytes = 326548688
sha256 = "4918ca09e05e41d2b4aa1ace6201d1cd911ffc58a42801002bab177d495cfe0a"
licence = "MIT"
upstream = "https://huggingface.co/ResembleAI/chatterbox-turbo-ONNX/resolve/d21799bd0354adb85e348b8a0442a8405110a2cf/onnx/conditional_decoder_quantized.onnx_data"

[[model]]
id = "chatterbox-turbo-speech_encoder_quantized.onnx_data"
family = "chatterbox"
file = "chatterbox-turbo-speech_encoder_quantized.onnx_data"
bytes = 354676576
sha256 = "d59861fb55e806fbeee731da9d4f8ff819fb5735de5d15e262d902594ee4dbb6"
licence = "MIT"
upstream = "https://huggingface.co/ResembleAI/chatterbox-turbo-ONNX/resolve/d21799bd0354adb85e348b8a0442a8405110a2cf/onnx/speech_encoder_quantized.onnx_data"

[[model]]
id = "chatterbox-turbo-language_model_quantized.onnx_data"
family = "chatterbox"
file = "chatterbox-turbo-language_model_quantized.onnx_data"
bytes = 367962860
sha256 = "ec9945df36cb5d131d46688f2609fd715fbfcb0b8ee9681af5c84118de2d55a2"
licence = "MIT"
upstream = "https://huggingface.co/ResembleAI/chatterbox-turbo-ONNX/resolve/d21799bd0354adb85e348b8a0442a8405110a2cf/onnx/language_model_quantized.onnx_data"

# ── Transcription ────────────────────────────────────────────────────────
# src/crates/concat-speech/src/transcribe.rs. whisper.cpp's GGML
# conversions of OpenAI's whisper.

[[model]]
id = "whisper-tiny.en"
family = "whisper"
file = "ggml-tiny.en.bin"
bytes = 77704715
sha256 = "921e4cf8686fdd993dcd081a5da5b6c365bfde1162e72b08d75ac75289920b1f"
licence = "MIT"
upstream = "https://huggingface.co/ggerganov/whisper.cpp/resolve/5359861c739e955e79d9a303bcbc70fb988958b1/ggml-tiny.en.bin"

[[model]]
id = "whisper-tiny"
family = "whisper"
file = "ggml-tiny.bin"
bytes = 77691713
sha256 = "be07e048e1e599ad46341c8d2a135645097a538221678b7acdd1b1919c6e1b21"
licence = "MIT"
upstream = "https://huggingface.co/ggerganov/whisper.cpp/resolve/5359861c739e955e79d9a303bcbc70fb988958b1/ggml-tiny.bin"

[[model]]
id = "whisper-base.en"
family = "whisper"
file = "ggml-base.en.bin"
bytes = 147964211
sha256 = "a03779c86df3323075f5e796cb2ce5029f00ec8869eee3fdfb897afe36c6d002"
licence = "MIT"
upstream = "https://huggingface.co/ggerganov/whisper.cpp/resolve/5359861c739e955e79d9a303bcbc70fb988958b1/ggml-base.en.bin"

[[model]]
id = "whisper-base"
family = "whisper"
file = "ggml-base.bin"
bytes = 147951465
sha256 = "60ed5bc3dd14eea856493d334349b405782ddcaf0028d4b5df4088345fba2efe"
licence = "MIT"
upstream = "https://huggingface.co/ggerganov/whisper.cpp/resolve/5359861c739e955e79d9a303bcbc70fb988958b1/ggml-base.bin"

[[model]]
id = "whisper-small.en"
family = "whisper"
file = "ggml-small.en.bin"
bytes = 487614201
sha256 = "c6138d6d58ecc8322097e0f987c32f1be8bb0a18532a3f88f734d1bbf9c41e5d"
licence = "MIT"
upstream = "https://huggingface.co/ggerganov/whisper.cpp/resolve/5359861c739e955e79d9a303bcbc70fb988958b1/ggml-small.en.bin"

[[model]]
id = "whisper-small"
family = "whisper"
file = "ggml-small.bin"
bytes = 487601967
sha256 = "1be3a9b2063867b937e64e2ec7483364a79917e157fa98c5d94b5c1fffea987b"
licence = "MIT"
upstream = "https://huggingface.co/ggerganov/whisper.cpp/resolve/5359861c739e955e79d9a303bcbc70fb988958b1/ggml-small.bin"