Text Match Cut
Type a sentence and generate a Vox-style archival newspaper match-cut montage. An LLM node splits the sentence into words, GPT Imagen 2 renders each word on an aged editorial page, and the frames assemble into a video. Adjust the frame rate and repeat count to dial in the stop-motion pacing, or edit the base prompt in the Chat Config node to shift the style (magazine colour, tabloid, ransom-note, comic book, etc).
简介
This ComfyUI workflow turns any sentence into a Vox-style archival newspaper match-cut montage. An OpenAIChatNode (driven by OpenAIChatConfig) takes your sentence and splits it into a clean list of words. FL_PromptMulti then expands those words into per-word image prompts, and OpenAIGPTImage1 calls OpenAI’s gpt-image-1 model to render each word as a bold editorial headline on an aged newsprint page. SaveImage can optionally write out the individual frames for inspection and iteration.
你需要提供
- 正在读取工作流
在 ComfyUI 中运行。 此工作流无法在本页面运行。你可以在 Comfy Cloud 中打开（已预先加载所需内容），或下载后在自己的 ComfyUI 中使用。
在 Cloud 中试用常见问答
The OpenAIGPTImage1 node uses OpenAI’s gpt-image-1. Style is controlled by the base prompt in OpenAIChatConfig and the FL_PromptMulti template. Edit those to switch from aged newspaper to magazine color, tabloid, ransom-note collage, or comic-book aesthetics.
Use ImageBatchRepeatInterleaving to set how many times each frame is repeated, and adjust FPS in CreateVideo. More repeats or lower FPS increases on-screen time per word. For fast, punchy cuts, use fewer repeats and higher FPS.
Yes. In OpenAIChatConfig, instruct the chat node to return an array of meaningful words only (e.g., remove articles like “a,” “an,” “the,” and strip punctuation). You can also instruct it to uppercase or normalize each word for consistent typography.
Refine the image prompt in FL_PromptMulti to be explicit: specify a bold serif or sans-serif headline, centered layout, high-contrast ink on off-white newsprint, and minimal background clutter. Keeping each word short and using uppercase often improves results.
Each word triggers an image generation call to gpt-image-1, so long sentences increase time and API cost. Consider capping the word count in OpenAIChatConfig, batching test runs at smaller image sizes, and previewing via SaveImage before exporting the final video.