Save 4 hours! We auto-setup your workflow! Free!

Drop your workflow.json — we handle every dependency, custom node, and model. Just open the link and run.

Auto-Setup Workflow Json (Free) Now!
ComfyUI > Nodes > FunUI > FunUI Guide Frames

ComfyUI Node: FunUI Guide Frames

Class Name

FunUIGuideFrames

Category
FunUI/Conditioning
Author
digital-garbage (Account age: 1276days)
Extension
FunUI
Latest Updated
2026-11-03
Github Stars
0.02K

How to Install FunUI

Install this extension via the ComfyUI Manager by searching for FunUI
  • 1. Click the Manager button in the main menu
  • 2. Select Custom Nodes Manager button
  • 3. Enter FunUI in the search bar
After installation, click the Restart button to restart ComfyUI. Then, manually refresh your browser to clear the cache and access the updated list of nodes.

Visit ComfyUI Online for ready-to-use ComfyUI environment

  • Free trial available
  • 16GB VRAM to 80GB VRAM GPU machines
  • 400+ preloaded models/nodes
  • Freedom to upload custom models/nodes
  • 200+ ready-to-run workflows
  • 100% private workspace with up to 200GB storage
  • Dedicated Support

Run ComfyUI Online

FunUI Guide Frames Description

Applies reference images as LTX video guides to maintain visual continuity between clips.

FunUI Guide Frames:

FunUIGuideFrames helps maintain visual continuity between LTX video clips by placing pictures from your media library into a new shot as gentle visual guides. It uses ComfyUI’s native LTX guide-attention mechanism, allowing earlier images to influence the new generation without necessarily locking it to the previous composition. In FunUI, this supports two approaches: using the middle picture of the preceding clip to help preserve the room, props, and supporting characters, or using pictures from several earlier clips as visual memory. The memory approach requires the JoyAI-Echo LoRA because the base model was not trained to interpret those pictures as memory. FunUI chooses the pictures using its timeline information; the node applies the supplied selection rather than choosing earlier clips itself. This is an experimental continuity feature, and stronger guidance can carry more of the previous composition into the new shot. When no guide list is supplied, the node leaves the generation data unchanged.

FunUI Guide Frames Input Parameters:

positive

A required CONDITIONING connection containing the positive guidance for your generation, such as the encoded description of what you want the video to show. When guide pictures are available, the node passes this conditioning through ComfyUI’s LTX guide mechanism along with the pictures, so the subsequent generation can use both the existing guidance and the visual references. With an empty guide list, this connection passes through unchanged. This is connected workflow data rather than a numeric setting; no minimum, maximum, default value, or selectable options are specified.

negative

A required CONDITIONING connection containing the negative guidance used by your workflow, typically describing qualities or content you want to avoid. The node passes this conditioning through the same LTX guide operation as the positive conditioning, keeping both conditioning paths compatible with the guided latent sequence. It does not simply add a negative text prompt or treat the guide pictures as unwanted content. With an empty guide list, the original negative conditioning passes through unchanged. No minimum, maximum, default value, or selectable options are specified for this connection.

vae

A required VAE connection used by the LTX guide mechanism to encode each selected picture into the representation used during video generation. The node also reads the VAE’s temporal downscaling information to translate a position within the clip into a video-frame index. Use the compatible video VAE from your LTX workflow; an unrelated image VAE may not provide the expected information or compatible encoding. This is a model connection, so the interface specifies no numeric range, default value, or selectable options.

latent

A required LATENT connection containing the video-only latent sequence that will receive the guides. This must be the picture latent, not an already joined picture-and-sound latent. The node uses its temporal length, together with the VAE’s downscaling information, to determine the original clip’s frame count and place each guide at the requested relative position. The LTX guide mechanism modifies this latent sequence to carry the reference pictures during sampling. If your workflow combines audio and video, apply this node to the separated video branch before joining the branches again. No minimum, maximum, default value, or selectable options are declared for this connection.

frames

An optional STRING containing a JSON list of guide-picture entries; its default is an empty string, which leaves all three returned connections unchanged. The interface shows the format [`{"media_id": "...", "at": 0.5, "strength": 0.25}`, ...], and FunUI normally writes this list for you. Each entry identifies a picture in the media library through media_id, supplies its relative position within the clip through at, and sets its influence through strength. A position of 0.5 means the midpoint; the node calculates the frame index by multiplying the position by one less than the original clip’s pixel-frame count and rounding to the nearest integer. For example, [`{"media_id": "YOUR_LIBRARY_PICTURE_ID", "at": 0.5, "strength": 0.25}`] requests one midpoint guide. Higher strength can carry more of the earlier composition into the new shot. The node interface declares no minimum, maximum, or selectable options for this string or its JSON fields, and the supplied code does not expose the parser’s complete validation rules; FunUI’s separate settings panel recommends a strength of 0.25 and offers a strength slider from 0.25 to 1.0. Pictures that cannot be found are skipped with a warning, while invalid list data reported by the parser stops execution.

FunUI Guide Frames Output Parameters:

positive

The CONDITIONING connection returned after applying all available guide pictures through the LTX guide mechanism. Connect this to the positive conditioning path used for subsequent sampling so that the generation receives the guide-related conditioning together with your existing positive guidance. This is conditioning data, not a generated image or a rewritten text prompt. If the guide list is empty, the original positive conditioning is returned unchanged.

negative

The CONDITIONING connection returned after processing the guide pictures through the negative conditioning path of the LTX guide mechanism. Connect this to the negative conditioning path used for subsequent sampling, alongside the corresponding positive output. Its purpose is to carry the conditioning produced by the same guide operations, rather than to describe the pictures as things to avoid. If the guide list is empty, the original negative conditioning is returned unchanged.

latent

The LATENT connection containing the video sequence after available guide pictures have been added. This remains generation data that must be sampled and decoded, not a finished video. Guides occupy additional sequence space during sampling, so remove them afterward with LTXVCropGuides before decoding the final video. In an audio-video workflow, this output is the video branch to reconnect with the audio branch as appropriate for your pipeline. With an empty guide list, the exact original latent is returned unchanged.

FunUI Guide Frames Usage Tips:

  • Start with a guide strength of 0.25. FunUI’s supplied notes identify this as the recommended balance and practical floor: higher values can pull the new shot toward the previous composition, while lower values were associated with noisy sound in its tests because audio still attends to the guide-containing sequence.
  • For continuity between adjacent shots, use FunUI’s “Middle frame of the clip before” mode. This selects a picture from the preceding clip to help preserve static surroundings and supporting elements while allowing the action to change; it is an app setting, not an additional input on this node.
  • Use “Memory from the last clips” only with the JoyAI-Echo LoRA loaded. FunUI defaults to remembering three clips and allows one to eight in its settings panel; these choices determine the list sent to the node rather than adding node inputs.
  • Keep the node on the video-only branch. In an LTX audio-video pipeline, separate the audio and video latents, apply any initial-image anchor and then the guides to the video latent, and reconnect the audio branch afterward.
  • Use all three outputs in the downstream generation path. The guide operation updates conditioning as well as latent data, so forwarding only the latent does not preserve the complete result of the node.
  • Add LTXVCropGuides after sampling and before decoding. Guide pictures sit at the tail of the sequence during sampling and should not be decoded as part of the finished shot.
  • When supplying the JSON manually, use real media-library picture IDs rather than assuming a file path will work. Start with one entry at 0.5 and strength 0.25 to check that the picture resolves and the workflow is connected correctly.
  • Leave frames empty when you do not need continuity guidance. This provides a true pass-through and avoids adding guide pictures; FunUI’s settings hint estimates that enabled guidance can cost roughly half again as much per sampling step, though this is not a fixed performance guarantee.

FunUI Guide Frames Common Errors and Solutions:

FunUI Guide Frames: …

  • Explanation: This is the error prefix used when the guide-list parser reports invalid data. The node appends the parser’s explanation and a final period, but the supplied context does not show the exact validation messages or all accepted limits.
  • Solution: Check that frames contains a valid JSON list with entries using media_id, at, and strength. Use double quotes for JSON keys and string values, remove trailing commas, and begin with the example format from the input description. If FunUI normally supplies the list, try disabling and re-enabling the guide setting or reselecting the source pictures.

a guide picture is gone from the library and was left out (MEDIA_ID)

  • Explanation: This is a warning rather than a fatal error. The requested picture could not be read from the media library, so the node skips it and continues with any other available guides. If every picture is missing, no guides are applied.
  • Solution: Check that the picture still exists in your media library and that the list refers to its current media ID. Restore or reselect the picture, or remove the outdated entry from the guide list.

Incompatible latent or VAE data

  • Explanation: This is a possible integration failure, not a fixed error message defined by this node. Execution expects a latent with video samples and a compatible VAE exposing temporal downscaling information. A joined audio-video latent, an unrelated VAE, or malformed latent data can fail during frame-count calculation or guide encoding.
  • Solution: Connect the LTX video-only latent and the matching VAE from the same workflow. In an audio-video pipeline, separate the branches before this node and join them again afterward. Use the complete underlying error message to identify which connection failed.

LTXVAddGuide cannot be imported or executed

  • Explanation: The node loads ComfyUI’s native LTXVAddGuide implementation when a nonempty guide list is processed. If that implementation is unavailable or incompatible with the installed environment, execution can fail; the exact exception depends on the installation.
  • Solution: Verify that your ComfyUI installation includes LTXVAddGuide and that your FunUI and ComfyUI versions are compatible. Update compatible components if necessary and restart ComfyUI. An empty guide list bypasses this import, so successful pass-through alone does not confirm that guide processing is available.

Guide pictures appear at the end of the decoded video

  • Explanation: This is a workflow symptom rather than an error message. The guide mechanism places additional guide content at the tail of the sequence during sampling, and decoding without removing it can include that content in the result.
  • Solution: Run LTXVCropGuides after sampling and before decoding. Follow the crop placement used by your LTX pipeline, especially when the sampled sequence contains both audio and video.

Memory guidance has little effect, or the previous composition dominates

  • Explanation: These are visual-result issues rather than runtime errors. Memory guidance requires the JoyAI-Echo LoRA, while excessive guide strength can carry too much of the earlier layout into a new shot. Neither continuity approach guarantees that every character, prop, or action will remain unchanged.
  • Solution: Confirm that the JoyAI-Echo LoRA is loaded when using Memory mode. Start at strength 0.25, choose pictures relevant to the new shot, and increase strength cautiously. For simpler continuity with the immediately preceding clip, try Middle frame mode instead.

Audio becomes noisy when using very weak guides

  • Explanation: The supplied FunUI test notes report that strengths below 0.25 can leave sound as noise because guide content remains physically present in the sequence and audio attends to it regardless of its strength. This is a reported behavior, not a dedicated exception raised by the node.
  • Solution: Use 0.25 as the starting strength rather than reducing it below the documented floor. If you do not need guidance, disable it or leave frames empty instead of attempting to neutralize it with an extremely low strength.

FunUI Guide Frames Related Nodes

Go back to the extension to check out more related nodes.
FunUI
RunComfy
Copyright 2025 RunComfy. All Rights Reserved.

RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.