comfy_scenes_json_node Introduction
comfy_scenes_json_node is a collection of ComfyUI nodes for working with scene-based creative workflows. The supplied source shows tools for making temporary images, videos, text, and audio; generating local narration and music; saving work while passing it to the next node; loading images by name; and gathering a run’s outputs into one folder.
These tools can help you build and check a story workflow in stages. For example, you can use mock nodes to test how scenes connect before spending time generating final media. You can then create narration or background music, save intermediate results for review, and collect the files produced by a completed run.
The supplied context does not include an upstream README, a complete workflow, or the implementation of every registered node. This guide describes only behavior visible in the provided source.
How comfy_scenes_json_node Works
Think of a ComfyUI workflow as a sequence of workstations. One node produces an image, piece of text, audio clip, or video; another node receives it and performs the next task. This extension supplies several kinds of workstation:
- Mock nodes produce quick stand-ins. They let you check connections and timing without claiming to make finished artwork or narration.
- Local audio nodes turn text into spoken audio or a music description into background music.
- Save passthrough nodes write a copy to ComfyUI’s output folder and return the same content, so the workflow can continue.
- File-management nodes find existing images or organize outputs after a run.
A batch means several items travel through a node together. For instance, Load Images By Names can return several matched images as one batch. If their dimensions differ, it resizes later images to match the first.
comfy_scenes_json_node Features
Test a workflow with mock content
Mock Scene Image makes a seeded, colored gradient with slight noise. It accepts a prompt, seed, quality, background, size, image count (n), and model selection, plus optional image and mask inputs. In the supplied implementation, the prompt and model selection do not make it generate prompt-matched artwork: the visible result is the synthetic gradient. Changing the seed changes its appearance; changing n produces more images. Its auto size resolves to 768 × 1344 pixels.
Mock Scene Video produces a 24-frame-per-second video with a gentle brightness change. It starts from the first frame of reference_0 when supplied, otherwise reference_1, or a neutral gray frame if neither is connected. Resolution, aspect ratio, and duration determine the output dimensions and frame count. Its prompt and model selection do not turn it into a generative video node.
Mock Scene Audio makes a tone, not spoken words. Its approximate length follows the amount of text and the speed setting. Voice and pitch affect the tone’s register. Empty text produces half a second of silence. It has the same audio inputs and output format as Kokoro Scene Audio, making it useful when checking the narration part of a workflow before generating speech.
Mock Scene Text returns the text you enter in mock_response. If that field is empty, it returns a diagnostic string describing inputs such as the prompt, selected model, image count, and file count. It does not call the model named in its selector.
Generate local narration
Kokoro Scene Audio (local TTS) converts text into spoken audio using a local Kokoro pipeline. Choose a voice, adjust speed from 0.5 to 2.0, and shift pitch from −12 to +12 semitones. Positive pitch values raise the voice; negative values deepen it. The optional connected voice_name overrides the voice dropdown. The node accepts a Kokoro voice ID directly and also passes other names through its voice-name resolver.
Voice IDs beginning with af_ or am_ are American English voices; bf_ and bm_ denote British English voices. The second letter indicates the voice category shown in the source: f or m. For example, af_sky and bm_george select different voices and accents. This node outputs 24 kHz mono audio. If the text is empty, it outputs half a second of silence instead of speech.
Generate local background music
MusicGen Audio (local music) creates audio from a written music description. A prompt such as “soft piano and gentle strings” describes the desired sound; prompt_override, when connected and nonempty, takes priority over the prompt field.
You can choose a duration from 1 to 300 seconds and adjust:
guidance_scale: how strongly generation follows the prompt.temperatureandtop_k: sampling controls that affect variation.seed: a value for repeatable sampling settings.fade_out_seconds: a gradual ending, up to 10 seconds.keep_model_loaded: keeps the model ready for another use, at the cost of continuing to occupy memory.
The node generates audio in windows of up to 30 seconds and uses the end of one window to continue longer pieces. It returns both the audio and its resulting duration. Longer pieces may take more time to produce.
Save work without interrupting the workflow
Save Image Passthrough, Save Audio Passthrough, and Save Text Passthrough write files under ComfyUI’s output directory and return their original input for use by the next node. Images are saved as PNG, audio as WAV, and text as UTF-8 text using the extension you specify (txt by default).
Each node lets you set a filename_prefix, an output_folder, and add_counter. With the counter enabled, filenames receive a numbered suffix such as _00001, and the node looks for a free number. With it disabled, a later run using the same name can replace an existing file. You can also include a subfolder in the filename prefix. Save Video Passthrough is registered by the extension, but its implementation is truncated in the supplied context, so its exact saving behavior is not documented here.
Reuse images and organize results
Load Images By Names searches a folder for image filenames matching the names you provide, without requiring file extensions in those names. It accepts a list, a JSON-style list, or comma- or newline-separated names. Matching is case-insensitive and supports PNG, JPG, JPEG, WebP, TGA, and BMP. A relative folder is interpreted beneath ComfyUI’s output directory.
The node returns the images, masks, a match count, and the matched paths. Missing names are skipped. If nothing matches, it returns a neutral gray 512 × 512 placeholder and a count of 0. Check the count: the placeholder keeps a workflow supplied with an image, but it is not one of your requested references.
Collect Run Outputs moves loose files and subfolders in ComfyUI’s output directory into a dated, numbered run folder, such as 17.11.2026 001. Existing folders that match its run-archive naming pattern are skipped. You can customize the date format and number padding, or enable dry_run to see what it would move without moving anything. Connect a value to its required trigger input to place it at the end of a workflow. Be aware that it collects all loose items in the output directory, not only files made by a particular scene.
The extension also registers Story Final Compile, YouTube Trend Topics, Video Last Frame, String → ElevenLabs Voice, and String → Kokoro Voice. Their implementations are not included in the supplied context, so this guide cannot reliably describe their controls or results.
comfy_scenes_json_node Models
The model names below are taken from the visible node choices. A name in a mock node’s dropdown does not mean that node runs the named service or model.
| Node | Available choices | What the choice means here |
|---|---|---|
| Kokoro Scene Audio | Kokoro voices including af_heart, af_sky, am_michael, bf_lily, and bm_george | Selects the voice used for local text-to-speech. |
| MusicGen Audio | facebook/musicgen-small, facebook/musicgen-medium, facebook/musicgen-stereo-small, facebook/musicgen-stereo-medium | Selects the local music model. small is the lighter option; medium is larger. stereo choices produce stereo rather than mono output. |
| Mock Scene Image | gpt-image-1.5, gpt-image-1, mock | Labels a mock run; all choices use the visible synthetic-image code. |
| Mock Scene Video | grok-imagine-video, mock | Labels a mock run; neither choice invokes generative video in the supplied implementation. |
| Mock Scene Text | gpt-5.5-pro, gpt-5.5, gpt-5-mini, gpt-4.1, gpt-4o, mock | Appears in the returned diagnostic text when no mock_response is provided; it does not generate a model response. |
For a quick workflow check, choose the mock nodes. For actual local narration, use Kokoro Scene Audio; for generated background music, use MusicGen Audio.
Troubleshooting comfy_scenes_json_node
An image, video, or text result does not follow my prompt.
- Check whether the node name starts with Mock.
- If it does, treat its result as a workflow test. The visible mock image and video implementations do not interpret prompts to create scenes, and Mock Scene Text uses
mock_responseor diagnostic text. - Use a separate, appropriate generation node when you need finished prompt-based visuals or writing.
Load Images By Names returns a gray image.
- Check its
countoutput.0means no requested files matched. - Confirm that
folderpoints to the directory containing your images. A relative path starts from ComfyUI’s output directory. - Check that each requested name matches a filename without its extension. For example, request
lunaforluna.png. - Check that the files use one of the supported image extensions.
Narration is silent, or the mock audio is not speech.
- Check that the connected
textcontains non-whitespace characters. Both the Kokoro and mock audio nodes return a short silence for empty text. - If you used Mock Scene Audio, switch to Kokoro Scene Audio when you need spoken narration. The mock produces a test tone by design.
- If a connected
voice_nameis present, remember that it overrides the Kokoro voice dropdown.
Saved files are hard to find, or a file was replaced.
- Look in ComfyUI’s output directory and the
output_folderyou specified. - Check whether
filename_prefixitself contains a subfolder path. - Enable
add_counterwhen you want the save node to choose a free numbered filename. Without it, the same name can be written again.
Collect Run Outputs moved more files than expected.
- Use
dry_runto preview the next collection. - Check the top level of ComfyUI’s output directory before running the node: it moves loose files and non-archive subfolders there.
- Keep unrelated loose outputs elsewhere before triggering a collection.
Frequently asked questions
- Do the mock nodes call the models shown in their dropdowns? No. The supplied mock implementations generate stand-ins locally or return diagnostic text.
- Can I use the same audio connection when switching from Mock Scene Audio to Kokoro Scene Audio? The supplied source gives them matching inputs and the same audio output shape for that purpose.
- What happens when only some requested images are found? The found images are returned in requested order; missing names are skipped, and
countreports how many were loaded. - Does
dry_runcreate an archive? No. It reports the proposed folder and moves in the log without creating or moving the archive contents.
