Skip to content

FunASR-powered video transcription, subtitle generation, and LLM-assisted clipping tool with a local Gradio UI.

feat(llm): add Cheaper Inference as a named OpenAI-compatible gateway (#223)

aiapienthusiast36acf72e2add
diff --git a/README.md b/README.md
index 65ded3f..c6ba6b4 100644
--- a/README.md
+++ b/README.md
@@ -138,6 +138,12 @@ OrcaRouter exposes one endpoint for all frontier and open-weight models, so you
Set `ORCAROUTER_API_KEY` (and optionally `ORCAROUTER_API_BASE`, which defaults to `https://api.orcarouter.ai/v1`) instead of pasting the key into the UI if you prefer environment-based configuration. A key is available at https://www.orcarouter.ai.
+#### Using Cheaper Inference as your LLM gateway (optional)
+
+FunClip can also send LLM-assisted clipping through [Cheaper Inference](https://cheaperinference.com), an OpenAI-compatible LLM gateway. One API key gives access to models from several labs. Select any `cheaperinference/` model in the **LLM Model Name** dropdown, paste a Cheaper Inference API key in the **APIKEY** box, and click 'LLM Inference'. FunClip removes the `cheaperinference/` prefix and sends the transcript and prompts to `https://api.cheaperinference.com/v1/chat/completions`. The returned segments work with the existing 'AI Clip' button.
+
+Set `CHEAPER_INFERENCE_API_KEY` (and optionally `CHEAPER_INFERENCE_API_BASE`, which defaults to `https://api.cheaperinference.com/v1`) instead of pasting the key into the UI if you prefer environment-based configuration. A key is available at https://cheaperinference.com/signup.
+
#### Content-aware clipping with TwelveLabs Pegasus (optional)
Besides the transcript-based LLMs above, FunClip can optionally use [TwelveLabs](https://twelvelabs.io) Pegasus, a video understanding model that reasons over the actual video (visuals + audio) rather than only the ASR transcript. This helps pick highlight segments even when the transcript alone is ambiguous (e.g. action, scene changes, on-screen events). To use it, select the `pegasus1.5` model name, paste your TwelveLabs API key, upload a video, and click 'LLM Inference' — Pegasus returns segments in the same `N. [start-end] text` format, so the existing 'AI Clip' button works unchanged. It needs `pip install twelvelabs`, and a free API key is available at https://twelvelabs.io.
diff --git a/README_zh.md b/README_zh.md
index 49a534a..7836eac 100644
--- a/README_zh.md
+++ b/README_zh.md
@@ -163,6 +163,12 @@ OrcaRouter 用单一端点接入所有前沿与开源模型,无需修改 FunCl
也可以不填 UI,而是设置 `ORCAROUTER_API_KEY` 环境变量(可选 `ORCAROUTER_API_BASE`,默认为 `https://api.orcarouter.ai/v1`)。Key 可在 https://www.orcarouter.ai 获取。
+#### 使用 Cheaper Inference 作为 LLM 网关(可选)
+
+FunClip 也可以将 LLM 智能裁剪发送到 [Cheaper Inference](https://cheaperinference.com)——一个 OpenAI 兼容的 LLM 网关。一个 API key 即可使用多家厂商的模型。在 **LLM Model Name** 下拉框选择任意 `cheaperinference/` 模型,在 **APIKEY** 输入框粘贴 Cheaper Inference API key,点击“LLM推理”。FunClip 会去掉 `cheaperinference/` 前缀,把字幕与 prompt 发送到 `https://api.cheaperinference.com/v1/chat/completions`,返回的分段与现有“AI Clip”按钮兼容。
+
+也可以不填 UI,而是设置 `CHEAPER_INFERENCE_API_KEY` 环境变量(可选 `CHEAPER_INFERENCE_API_BASE`,默认为 `https://api.cheaperinference.com/v1`)。Key 可在 https://cheaperinference.com/signup 获取。
+
### B.通过命令行调用使用FunClip的相关功能
```shell
# 下载下面命令用到的示例视频
diff --git a/funclip/launch.py b/funclip/launch.py
index cfca30d..08ae700 100644
--- a/funclip/launch.py
+++ b/funclip/launch.py
@@ -150,7 +150,7 @@ if __name__ == "__main__":
)
def llm_inference(system_content, user_content, srt_text, model, apikey, video_input=None):
- SUPPORT_LLM_PREFIX = ['litellm', 'qwen', 'gpt', 'g4f', 'moonshot', 'deepseek', 'atlascloud', 'minimax', 'orcarouter', 'pegasus']
+ SUPPORT_LLM_PREFIX = ['litellm', 'qwen', 'gpt', 'g4f', 'moonshot', 'deepseek', 'atlascloud', 'minimax', 'orcarouter', 'cheaperinference', 'pegasus']
if model.startswith('litellm/'):
return litellm_call(apikey, model, user_content+'\n'+srt_text, system_content)
if model.startswith('pegasus'):
@@ -162,7 +162,7 @@ if __name__ == "__main__":
return call_twelvelabs_pegasus(apikey, video_input, model=model, prompt=system_content)
if model.startswith('qwen'):
return call_qwen_model(apikey, model, user_content+'\n'+srt_text, system_content)
- if model.startswith('gpt') or model.startswith('moonshot') or model.startswith('deepseek') or model.startswith('atlascloud/') or model.startswith('minimax/') or model.startswith('orcarouter/'):
+ if model.startswith('gpt') or model.startswith('moonshot') or model.startswith('deepseek') or model.startswith('atlascloud/') or model.startswith('minimax/') or model.startswith('orcarouter/') or model.startswith('cheaperinference/'):
return openai_call(apikey, model, user_content+'\n'+srt_text, system_content)
elif model.startswith('g4f'):
model = "-".join(model.split('-')[1:])
@@ -273,6 +273,10 @@ if __name__ == "__main__":
"orcarouter/fusion",
"orcarouter/fusion-flash",
"orcarouter/fusion-mini",
+ "cheaperinference/gpt-5.4-mini",
+ "cheaperinference/gpt-5.4",
+ "cheaperinference/claude-sonnet-5",
+ "cheaperinference/deepseek-v4-flash",
"pegasus1.5"],
value="deepseek-chat",
label="LLM Model Name",
diff --git a/funclip/llm/openai_api.py b/funclip/llm/openai_api.py
index 4882558..66615df 100644
--- a/funclip/llm/openai_api.py
+++ b/funclip/llm/openai_api.py
@@ -19,6 +19,11 @@ MINIMAX_MODEL_PREFIX = "minimax/"
ORCAROUTER_API_BASE = "https://api.orcarouter.ai/v1"
ORCAROUTER_MODEL_PREFIX = "orcarouter/"
+# Cheaper Inference is an OpenAI-compatible LLM gateway. Its model IDs are
+# bare (e.g. `gpt-5.4-mini`), so the `cheaperinference/` prefix is stripped.
+CHEAPER_INFERENCE_API_BASE = "https://api.cheaperinference.com/v1"
+CHEAPER_INFERENCE_MODEL_PREFIX = "cheaperinference/"
+
def _resolve_model_config(model):
base_url = None
@@ -53,6 +58,16 @@ def _resolve_model_config(model):
if not base_url:
base_url = ORCAROUTER_API_BASE
api_key_env = "ORCAROUTER_API_KEY"
+ elif model.startswith(CHEAPER_INFERENCE_MODEL_PREFIX):
+ model = model[len(CHEAPER_INFERENCE_MODEL_PREFIX):]
+ if not model:
+ raise ValueError(
+ "Model name is empty after stripping cheaperinference/ prefix"
+ )
+ base_url = os.environ.get("CHEAPER_INFERENCE_API_BASE", CHEAPER_INFERENCE_API_BASE).strip()
+ if not base_url:
+ base_url = CHEAPER_INFERENCE_API_BASE
+ api_key_env = "CHEAPER_INFERENCE_API_KEY"
elif model.startswith("deepseek"):
base_url = "https://api.deepseek.com"
elif model.startswith("gpt-3.5-turbo"):
diff --git a/tests/test_cheaperinference_api.py b/tests/test_cheaperinference_api.py
new file mode 100644
index 0000000..f142f9c
--- /dev/null
+++ b/tests/test_cheaperinference_api.py
@@ -0,0 +1,92 @@
+"""Tests for Cheaper Inference routing through the OpenAI-compatible client."""
+
+import os
+import unittest
+from unittest.mock import MagicMock, patch
+
+from funclip.llm.openai_api import (
+ CHEAPER_INFERENCE_API_BASE,
+ openai_call,
+)
+
+
+def _mock_completion(content="ok"):
+ completion = MagicMock()
+ completion.choices = [MagicMock()]
+ completion.choices[0].message.content = content
+ return completion
+
+
+class TestCheaperInferenceRouting(unittest.TestCase):
+ def test_cheaperinference_prefix_uses_gateway_base_url(self):
+ client = MagicMock()
+ client.chat.completions.create.return_value = _mock_completion("clip plan")
+
+ with patch("funclip.llm.openai_api.OpenAI", return_value=client) as openai_cls:
+ result = openai_call(
+ "ci-key",
+ "cheaperinference/gpt-5.4-mini",
+ "subtitle text",
+ "find highlights",
+ )
+
+ self.assertEqual(result, "clip plan")
+ openai_cls.assert_called_once_with(
+ api_key="ci-key",
+ base_url=CHEAPER_INFERENCE_API_BASE,
+ )
+ # Cheaper Inference model IDs are bare: the prefix is stripped.
+ call_kwargs = client.chat.completions.create.call_args[1]
+ self.assertEqual(call_kwargs["model"], "gpt-5.4-mini")
+
+ def test_cheaperinference_api_key_falls_back_to_env(self):
+ client = MagicMock()
+ client.chat.completions.create.return_value = _mock_completion()
+
+ with patch.dict(os.environ, {"CHEAPER_INFERENCE_API_KEY": "env-ci-key"}, clear=False):
+ with patch("funclip.llm.openai_api.OpenAI", return_value=client) as openai_cls:
+ openai_call("", "cheaperinference/gpt-5.4-mini", "text")
+
+ openai_cls.assert_called_once_with(
+ api_key="env-ci-key",
+ base_url=CHEAPER_INFERENCE_API_BASE,
+ )
+ call_kwargs = client.chat.completions.create.call_args[1]
+ self.assertEqual(call_kwargs["model"], "gpt-5.4-mini")
+
+ def test_cheaperinference_api_base_env_overrides(self):
+ client = MagicMock()
+ client.chat.completions.create.return_value = _mock_completion()
+
+ with patch.dict(
+ os.environ,
+ {"CHEAPER_INFERENCE_API_BASE": "https://gateway.example.com/v1"},
+ clear=False,
+ ):
+ with patch("funclip.llm.openai_api.OpenAI", return_value=client) as openai_cls:
+ openai_call("ci-key", "cheaperinference/claude-sonnet-5", "text")
+
+ openai_cls.assert_called_once_with(
+ api_key="ci-key",
+ base_url="https://gateway.example.com/v1",
+ )
+
+ def test_empty_cheaperinference_model_raises(self):
+ with self.assertRaises(ValueError):
+ openai_call("key", "cheaperinference/", "text")
+
+ def test_missing_cheaperinference_key_does_not_fall_back_to_openai_key(self):
+ with patch.dict(
+ os.environ,
+ {"OPENAI_API_KEY": "openai-only-key"},
+ clear=True,
+ ):
+ with patch("funclip.llm.openai_api.OpenAI") as openai_cls:
+ with self.assertRaisesRegex(ValueError, "CHEAPER_INFERENCE_API_KEY"):
+ openai_call("", "cheaperinference/gpt-5.4-mini", "text")
+
+ openai_cls.assert_not_called()
+
+
+if __name__ == "__main__":
+ unittest.main()
diff --git a/tests/test_cheaperinference_launch_integration.py b/tests/test_cheaperinference_launch_integration.py
new file mode 100644
index 0000000..03d1f52
--- /dev/null
+++ b/tests/test_cheaperinference_launch_integration.py
@@ -0,0 +1,74 @@
+"""Regression tests for Cheaper Inference choices and prompt routing in the launcher."""
+
+import ast
+import unittest
+from pathlib import Path
+
+
+LAUNCH_PATH = Path(__file__).resolve().parents[1] / "funclip" / "launch.py"
+
+
+class TestCheaperInferenceLaunchIntegration(unittest.TestCase):
+ @classmethod
+ def setUpClass(cls):
+ cls.tree = ast.parse(LAUNCH_PATH.read_text(encoding="utf-8"))
+
+ def test_openai_compatible_route_handles_cheaperinference_prefix(self):
+ llm_inference = next(
+ node
+ for node in ast.walk(self.tree)
+ if isinstance(node, ast.FunctionDef) and node.name == "llm_inference"
+ )
+ openai_call = next(
+ node
+ for node in ast.walk(llm_inference)
+ if isinstance(node, ast.Call)
+ and isinstance(node.func, ast.Name)
+ and node.func.id == "openai_call"
+ )
+ # The openai_call dispatch branch must also cover cheaperinference/ models.
+ dispatch_condition = next(
+ node
+ for node in ast.walk(llm_inference)
+ if isinstance(node, ast.If) and isinstance(node.test, ast.BoolOp)
+ )
+ self.assertIn("cheaperinference/", ast.unparse(dispatch_condition.test))
+ self.assertEqual(ast.unparse(openai_call.args[2]), "user_content + '\\n' + srt_text")
+ self.assertEqual(ast.unparse(openai_call.args[3]), "system_content")
+
+ def test_support_prefix_list_includes_cheaperinference(self):
+ llm_inference = next(
+ node
+ for node in ast.walk(self.tree)
+ if isinstance(node, ast.FunctionDef) and node.name == "llm_inference"
+ )
+ assigned = [
+ node
+ for node in ast.walk(llm_inference)
+ if isinstance(node, ast.Assign) and node.targets[0].id == "SUPPORT_LLM_PREFIX"
+ ]
+ self.assertEqual(len(assigned), 1)
+ prefix_list = assigned[0].value
+ self.assertIsInstance(prefix_list, ast.List)
+ prefixes = {
+ node.value
+ for node in ast.walk(prefix_list)
+ if isinstance(node, ast.Constant) and isinstance(node.value, str)
+ }
+ self.assertIn("cheaperinference", prefixes)
+
+ def test_dropdown_lists_cheaperinference_gateway_models(self):
+ string_literals = {
+ node.value
+ for node in ast.walk(self.tree)
+ if isinstance(node, ast.Constant) and isinstance(node.value, str)
+ }
+
+ self.assertIn("cheaperinference/gpt-5.4-mini", string_literals)
+ self.assertIn("cheaperinference/gpt-5.4", string_literals)
+ self.assertIn("cheaperinference/claude-sonnet-5", string_literals)
+ self.assertIn("cheaperinference/deepseek-v4-flash", string_literals)
+
+
+if __name__ == "__main__":
+ unittest.main()