AI News, 5 October: Photoshop meets ComfyUI, LTX meets post-production

01Two recent workflow demonstrations

A Photoshop bridge demonstration appeared on October 4, and LTX's VFX workflow recap appeared on October 2. They are useful looks at production handoffs, not October 5 launches. The bridge tackles the trip between a layered image and a ComfyUI graph. LTX's documentation tackles a different handoff: moving AI video results toward the mattes, dynamic range and output sizes a post-production team can inspect and composite.

02Photoshop Bridge: send an image job to your ComfyUI

Geekatplay Studio's Photoshop Bridge combines a Photoshop panel with ComfyUI nodes. A generation returns as a new layer; an edit sends a selected area or layer and places the result over its source area. The panel accepts bundled examples, saved workflows and imported workflow files. Its image inputs can come from selected layers, while the user's ComfyUI setup runs the graph.

Consider a poster with a photographed object, a separate background and approved typography. The artist could keep the text and object layers in Photoshop, send a background region to a chosen image-edit workflow, then compare the returned layer with the original before masking it in. For a second pass, separate reference layers could feed a multi-image graph. That is a proposed use case, not a test or a promise that the generated pixels will match an art director's brief. The useful handoff is the returned, inspectable layer: an artist can reject an unwanted change without flattening the whole poster.

Installation instructions cover Windows and macOS. You need Photoshop 2024 or later, Creative Cloud for the panel, a current local or reachable ComfyUI server, and the selected workflow's models. Photoshop documents can be 8- or 16-bit, but 32-bit is unsupported. A selection travels as its bounding rectangle, so check the returned edge on an irregular cutout. The bridge code is MIT licensed; Photoshop and model terms are separate.

03LTX: video handoffs for mattes, HDR and larger canvases

LTX's October 2 maker recap showed VFX-oriented workflows, but its official guidance is the better starting point for choosing one. The Alpha Gen beta guide describes a frame-aligned grayscale subject matte from RGB video. That matte is meant to travel alongside the original footage into a compositor; it is not a finished transparent beauty pass. A small team could try it on a short product shot, inspect the edge frame by frame, and use the matte to place graphics behind the subject. Fine hair, motion blur and occlusion would still need human attention. The guide lists a 121-frame recommendation, a 1920 by 1088 maximum, and no prompt or mask control for choosing which subject to isolate.

For a different problem, LTX's SDR-to-HDR guide describes a path to scene-linear EXR or BT.2020/HLG output. An editor could compare that converted clip with the SDR source in a managed color pipeline, then decide whether highlights and skin tones are usable for the intended grade. LTX explicitly says reconstructed highlight detail is synthesized, not recovered camera information. Its April 23 announcement and API changelog already documented SDR-to-HDR, so the October recap should be read as a workflow demonstration rather than the feature's first arrival. The demonstration discussed a newer version, but it did not establish a dated release for every component shown.

The Native Resolution guide describes a full-frame composition stage followed by a larger pass using overlapping Tiled Fusion regions on one latent canvas. A finishing artist could start with a short shot, compare the output against the approved edit, then inspect motion, fine edges, gradients and tile boundaries at full resolution. Full HD, 4K and 8K choices in the workflow are size selectors, not guarantees of speed, memory use or image quality. LTX says runtime and VRAM depend on the selected canvas and model stack, so a studio should test its own GPU and intended clip before planning a delivery around the highest setting.

These documented workflows call for local ComfyUI and matching model and adapter downloads. Keep LTX-2.3 and LTX-2.5 assets separate unless a specific workflow or model card establishes compatibility. LTX model cards carry community-license terms that need checking for the intended use. XANTRANS has not installed the bridge or LTX workflows, measured output quality, or cleared a model for a commercial job.