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 > comfui-helto-ltx

ComfyUI Extension: comfui-helto-ltx

Repo Name

comfui-helto-ltx

Author
helto4real (Account age: 3523 days)
Nodes
View all nodes(1)
Latest Updated
2026-07-03
Github Stars
0.0K

How to Install comfui-helto-ltx

Install this extension via the ComfyUI Manager by searching for comfui-helto-ltx
  • 1. Click the Manager button in the main menu
  • 2. Select Custom Nodes Manager button
  • 3. Enter comfui-helto-ltx 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

comfui-helto-ltx Description

comfui-helto-ltx adds image-guidance nodes for LTX 2.3 video in ComfyUI, placing stills or starting sequences at chosen frames with adjustable strength and reusable guides across generation stages.

comfui-helto-ltx Introduction

comfui-helto-ltx adds image-guidance nodes for creating LTX Video in ComfyUI. It is designed for LTX 2.3 workflows in which you want still images to influence particular moments in a video, such as its opening, a point in the middle, or its ending. You can also guide the beginning with a sequence of images, such as frames extracted from an existing clip.

The extension gives you two ways to work:

  • A compact workflow: Set up guides and apply them in one node, or use a generation node that produces decoded video frames and audio.
  • A reusable, multi-stage workflow: Set up your images once, then apply the same guide list separately to low-resolution and high-resolution generation stages.

This can save you from rebuilding an image list for every stage. It also gives you control over when each guide appears, how strongly it influences the result, and how images with different proportions fit the video frame.

The extension uses ComfyUI’s native LTXV guide behavior. Guides influence generation; they are not guaranteed to appear as exact copies of their source images.

How comfui-helto-ltx Works

Think of a guide image as a visual direction for one moment in your video. You choose an image, specify where it belongs on the timeline, and set its strength. During generation, the extension prepares that image in the format expected by ComfyUI’s native LTXV tools and passes the resulting guidance to the sampler.

A typical guided workflow has four parts:

  1. Choose images. Add individual still images, an optional starting image sequence, or both.
  2. Place them on the timeline. For example, position one image at frame 0 and another at -1, which means the final frame.
  3. Apply the guides. The extension resizes and encodes the images, then updates the video latent and its positive and negative conditioning.
  4. Sample and crop. The native LTXV guide process appends guide latents. In workflows where sampling is handled outside the extension’s generation node, use LTXVCropGuides after sampling when later nodes expect only the normal video frames.

A latent is the working representation of the video used during generation, before it is decoded into viewable frames. If you connect a latent made by another node, its dimensions and length take precedence over settings intended to create a new latent. This matters most when you work in two stages or troubleshoot an unexpected frame count.

For a two-stage workflow, the Image Guide Manager acts like a shared reference sheet: it holds your guide list and common settings. Each Apply Image Guides node then prepares those same references for its own stage’s resolution.

comfui-helto-ltx Features

Four nodes for different workflows

NodeWhat it doesWhen to use it
LTX 2.3 Image Guides (All-in-One)Combines guide setup and application. Outputs conditioning and a latent for a separate sampler.A compact single-stage workflow or an older workflow already using this node.
LTX 2.3 Image Guide ManagerStores the guide list and shared settings as a reusable IMAGE_GUIDES output. It does not encode guides on its own.A workflow with multiple sampling stages.
LTX 2.3 Apply Image GuidesApplies a Manager’s guide list to one sampling stage. Outputs updated conditioning and a latent.Use one per stage, such as one for low resolution and one for high resolution.
LTX 2.3 Generate All-in-OneEncodes prompts, prepares guides, samples, crops guide latents, decodes video, and returns an audio output.Single-pass generation when you want decoded frames and audio directly from one node.

For a two-stage workflow, connect one Manager to two Apply nodes. Give each Apply node the width, height, and optional input latent appropriate to its stage. The Manager’s width and height help with previews and aspect-ratio warnings; the Apply node’s dimensions control guide encoding for that stage.

Guide list and image browser

The custom guide interface is available on the Image Guides (All-in-One), Image Guide Manager, and Generate All-in-One nodes. You can add an image, add or remove a configured image folder, refresh image lists, and save or load a guide set.

Each guide row lets you enable or disable the image, set its position and strength, reorder it, or remove it. Hover over an image name to see a preview. The image browser also supports folder browsing, optional recursive subfolder listing, searching by filename or relative path, a thumbnail-visibility toggle, and a grid-column slider. Ctrl-click an image for a larger preview.

Saved guide sets can retain your guide items, folder aliases, timing mode, frame rate, size metadata, and shared settings. They are useful when you want to reuse a visual plan across workflows.

Timing and video length

You can enter guide positions as either frames or seconds:

  • frame mode: A position of 24 means approximately frame 24.
  • seconds mode: A position of 2 at 24 fps resolves to approximately frame 48.
  • Negative positions: -1 means the final pixel frame; -2 means the second-to-last.

Calculated positions are kept within the video’s frame range. Native LTXV video lengths should follow 8 × n + 1, such as 97, 105, or 113 frames. If the extension creates an empty latent and the length is invalid, it reports an error rather than silently producing fewer frames.

When you supply an existing latent, that latent determines the actual generated length. Keep its length in mind when positioning guides, especially an end guide.

Image fit and compression

The extension prepares guide images at the target stage’s resolution. Choose a resize mode according to the composition you want:

Resize modeEffect
containKeeps the whole image visible and fills unused space with padding. This is the default.
padThe same behavior as contain.
stretchFills the target size but may distort the image’s proportions.
cropFills the frame while preserving proportions, cropping overflow from the center.

For example, use contain when an entire character illustration must remain visible. Use crop when filling the video frame matters more than retaining its edges. The pad_color setting controls the padding color.

img_compression uses ComfyUI’s native LTXV image preprocessing before normal guide encoding. A value of 0 disables it; values from 1 to 100 enable it. The default is 35. Compression is not applied when an image is inserted as a locked latent frame, so that insertion remains more faithful to its prepared source.

Guide strength and overlapping positions

Each manual image has its own strength. Its effective strength is global_strength × per-image strength, limited to the range 0.0 to 1.0. A starting image sequence similarly uses global_strength × start_images_strength.

Increasing strength generally makes the result follow the image more closely, but can leave less freedom for motion. A guide alone does not create audio or lip sync; those depend on the rest of your video and audio workflow.

If two enabled manual guides land on the same frame, choose a duplicate_policy:

  • error: Stop and ask you to resolve the overlap.
  • keep_first: Use the earlier guide.
  • keep_last: Use the later guide.
  • offset_next: Move the later guide to the next available frame.

An overlap between a manual guide and the start image sequence is always an error; this policy does not override it.

Start sequences and stronger frame locking

The optional start_images input accepts a ComfyUI IMAGE batch, typically frames extracted from a video. It begins at frame 0 and is prepared as one native multi-frame guide. You can still add manual guides later in the timeline, such as an ending image at -1, provided their positions do not overlap the starting sequence.

By default, a manual guide at the beginning or end acts as a reference, not a hard replacement. Two optional settings provide stronger initialization:

  • lock_start_frames: Writes a frame-0 manual guide or the starting sequence into the beginning of the video latent.
  • lock_end_frame: Writes a final-frame manual guide into the final video latent slot.

Locked frames are still resized and VAE-encoded, so they are not pixel-perfect copies after decoding. End locking also affects the final latent block, not only one visible frame. If the ending looks distorted or the transition feels forced, turn off lock_end_frame and try a normal guide at -1.

Generation, schedules, and audio

Generate All-in-One accepts a model, CLIP, VAE, prompts, guide settings, and sampling controls such as seed, steps, CFG, and sampler. It outputs decoded video frames as IMAGE and audio as AUDIO. Unlike the other guide nodes, it performs sampling, guide cropping, and decoding internally. It is intended for single-pass workflows, not two-stage upscaling.

Its sigma_mode controls the sampling schedule:

  • ltx_scheduler: Uses the native LTXV scheduler with settings such as steps, shifts, stretch, and terminal.
  • manual: Uses the values entered in manual_sigmas. The effective number of denoising steps is one less than the number of values. In this mode, steps, stretch, and terminal are ignored; max_shift and base_shift still affect LTXV model sampling.

Its audio_mode controls the audio output:

  • passthrough: Samples video only. Connected audio is trimmed or padded to the video duration; without connected audio, the node outputs silence.
  • native_av: Requires an LTXV audio/video model and an audio_vae. Connected audio can be encoded as locked AV audio; without it, the node can create an empty audio latent and decode generated audio.

During a run, the node displays phase labels such as Encoding prompts, Preparing guides, Sampling video, Cropping guide latents, and Decoding video. The sampler continues to provide detailed progress during denoising.

comfui-helto-ltx Models

The extension does not provide its own models. It provides nodes that work with ComfyUI’s native LTXV components and the model inputs in your workflow.

  • For a guided video workflow, use an appropriate LTXV model with the native LTXV sampler path, or connect it to Generate All-in-One.
  • For native_av audio generation in Generate All-in-One, use an LTXV AV model and connect an audio VAE from the native LTXV Audio VAE Loader.
  • If you only need to carry an existing audio track alongside generated frames, choose passthrough instead of native_av.

The provided documentation does not identify specific model files or compare their quality and speed.

What’s New with comfui-helto-ltx

The provided documentation does not include a dated version history. It does describe these capabilities and changes, without assigning them to particular releases:

  • Reusable Manager/Apply workflow: One guide list can serve separate low- and high-resolution stages.
  • Generate All-in-One node: A single-pass option now covers prompt encoding, sampling, guide cropping, video decoding, and audio output.
  • Clearer invalid-length handling: Internally created empty latents report an error when the requested frame count does not fit the native LTXV length pattern.
  • Older-workflow compatibility: The original guide all-in-one node remains available, and Apply accepts legacy extra inputs at runtime to reduce migration failures.

No version numbers or release dates are supplied, so these items should not be read as a chronological changelog.

Troubleshooting comfui-helto-ltx

My video has the wrong number of frames

  1. Check that the requested native LTXV length follows 8 × n + 1, such as 97 or 105.
  2. If a latent is connected, inspect the node that creates it. Its length determines the actual generated length.
  3. Make sure the guide positions and Manager’s num_frames reflect that active length, particularly for guides near the end.

My last image does not appear at the end, or the ending looks strange

  1. Set the manual guide position to -1 for the final pixel frame.
  2. Check whether a connected latent has a different length from the Manager’s num_frames.
  3. If lock_end_frame is on and several ending frames look distorted or overly constrained, switch it off. Try the same image as a normal -1 guide for a less forceful transition.

The output has extra guide frames

  1. Find the video latent coming out of your sampler.
  2. Add the native LTXVCropGuides node after sampling.
  3. Send its cropped output to decoding or any later node that expects the normal video length.

Generate All-in-One performs this crop internally.

A guide conflicts with another image

  1. Check whether two enabled manual guides calculate to the same frame.
  2. Move one guide, disable one, or choose the appropriate duplicate_policy.
  3. If the error mentions start_images, move the manual guide outside the frames covered by that sequence. Start-sequence overlap cannot be resolved with duplicate_policy.

The Apply node shows old sockets

  1. Refresh the ComfyUI browser page.
  2. Restart ComfyUI if the sockets still look outdated.
  3. Recreate the Apply node in the workflow if its displayed interface remains stale.

Existing workflows can retain an older visual representation after a node schema changes.

Images, thumbnails, or a saved guide set are missing

  1. Use the guide toolbar’s Refresh folders/images action.
  2. Check that the configured folder alias still exists and points to the intended folder.
  3. Check that the image is still present and uses a supported extension: .png, .jpg, .jpeg, .webp, or .bmp.
  4. If loading a saved guide set, restore its missing folder alias or update the set. Workflows store aliases and filenames, not absolute paths.
  5. Restart ComfyUI if refreshed image lists still appear stale.

Generate All-in-One audio fails in native_av mode

  1. Confirm that the connected model supports LTXV audio/video generation.
  2. Connect an audio_vae from the native LTXV Audio VAE Loader.
  3. If you only want to keep an external audio track with the generated video, select passthrough.

Manual sigma parsing fails

  1. Select sigma_mode = manual.
  2. Enter at least two numeric sigma values in manual_sigmas.
  3. Remember that nine values produce eight denoising steps. In manual mode, changing the separate steps setting will not change that count.

Frequently Asked Questions

Do I need to use every node?
No. Use Image Guides (All-in-One) for a compact guide-and-sampler setup, Generate All-in-One for single-pass decoded video and audio, or Manager plus Apply when guides must be reused across stages.

Does half_size_first_pass resize a latent I connected?
No. It only affects an empty latent created by the node. A connected latent’s shape takes precedence.

Will locking a start or end image copy it exactly?
No. Locked images pass through resizing and VAE encoding and decoding. They provide stronger initialization than a normal guide, but do not guarantee pixel-identical frames.

Can image guides create sound or lip sync?
No. Guide strength controls visual influence. Sound and lip sync depend on the audio-capable model and the rest of your workflow.

Why does the extension store folder aliases?
Guide items and saved workflows refer to a folder alias and filename instead of embedding an absolute local path. The folder must remain configured for those references to work.

Learn More about comfui-helto-ltx

  • comfui-helto-ltx repository: The project source and its current documentation.
  • Native ComfyUI LTXV nodes: When examining a workflow, look for LTXVAddGuide and LTXVCropGuides. Understanding where guides enter the sampler and where appended guide latents are cropped makes it easier to troubleshoot the final video length.

comfui-helto-ltx Related Nodes

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.

comfui-helto-ltx detailed guide | ComfyUI