A small node pack with five purpose-built nodes. They have niche uses — I built them because I needed them for one of my workflows — but they work anywhere you need persistent frame caching, auto resolution without having to build a massive switch panel or mess around with the width and height settings yourself, grab first or last frames from multiple video/images, or multi-segment video assembly.PLEASE NOTE: these are experimental and im finding issues the more I mess with the nodes myself. I do testing myself and as I mess with them more I find something that I overlooked that ends up breaking workflows.If you find any bugs with these nodes please let me know as this is my first node pack and im still learning as I go. I will push an update through when ever I find issues myself, can find out some quality of life improvements, or add a new node.in version 4.0 I added started adding a change log with what issues I've run into and fixed so I can reference them as I go or if you get curious to see a more detailed change log than what is put in the description.Thank you all for trying out these nodes and thank you for your patience with this pack.<hr />📦 TrashAI Shelf Save and load image batches to disk. When images are connected, it saves the frames and passes them through. When bypassed or disconnected, it loads the last saved result automatically. Includes a lock toggle — lock a shelf to protect a result you like while you keep generating upstream. Data persists across restarts so you can work on one segment at a time without losing anything. Built for low VRAM workflows where you can't run everything at once.🖼 TrashAI Frame Picker Pick a single frame from up to 6 image or video batch inputs. Hit "Get Images" to extract the first and last frames from everything connected in one run. Then browse through them instantly using input_select and frame_mode — no re-running needed. When you've found the frame you want, toggle lock ON and hit run to pass it downstream. Designed to work alongside Shelf nodes for selecting i2v input frames without triggering the full pipeline or moving wires every time you switch.🔁 TrashAI Segment Repeater Combines up to 6 image batches into one continuous sequence. Connect your segments, set a repeat range, and it handles assembly with automatic seam trimming. Supports wraparound looping — set start=5 and end=2 to repeat S5→S6→S1→S2. Empty or disconnected slots are skipped automatically. Set repeat_count to 1 for straight assembly with no repeating. Repeated_images section works best if using a looped video segment. Works best with interpolation.🔗 TrashAI Segment Assembler Companion to Segment Repeater. Takes the before, repeated, and after outputs and concatenates them back into one sequence. All inputs are optional — connect what you need, skip what you don't.📐 TrashAI Resolution Selector Feed it any image and pick a target resolution — it resizes the image to match while keeping the aspect ratio intact and rounding both dimensions to the nearest multiple of 16. Scaling is based on the short side, so a 720x1280 portrait at 480p becomes 480x848. Set it to "auto" and it snaps to the closest standard preset based on whatever you feed it — no thinking required. Outputs the resized image plus width and height as integers. Designed for people who don't want to do math every time they change resolution, and for inputs that come in at weird non-standard sizes. Drop it before your I2V node or upscaler and forget about it.This node pack is a work in progress and will continue to get updates as I come up with more ideas. If these interest you I'd recommend subscribing so you get notified when I post a new version.This is only posted here on Civitai. If it appears on other websites it was uploaded without my knowledge or consent. This is the only location to get the official nodes.

TrashAI_nodes

SegmentRepeater<ul><li>Skips unconnected/placeholder inputs instead of crashing on tensor mismatch</li><li><code>repeat_start</code> and <code>repeat_end</code> clamp to whatever's actually real — set 1-6 once and forget</li><li>Gaps in connection order are handled silently</li></ul>ShelfNode<ul><li>Locking now skips upstream execution (proper buffer-zone behavior via <code>IS_CHANGED</code>)</li><li><code>torch.load</code> uses <code>weights_only=True</code> (PyTorch 2.6+ compatibility)</li><li>Won't overwrite saved data with placeholders</li><li>JS auto-rename no longer fires on workflow load (preserves slot names)</li></ul>FramePicker<ul><li>Locked mode uses cache properly — same selection reuses output, change selection re-runs</li><li>Locked mode raises clear error if selected input is empty (no silent fallback)</li><li>Preview filenames include node ID — no collisions between multiple Frame Pickers</li></ul>ResolutionSelector<ul><li>Output dimensions have a 16px floor to prevent zero-dim outputs</li></ul>SegmentAssembler — no functional changes

Files with SHA256: dd2c425026ac6f933720e690c2043f53859643984835fa1e43c9907fb9f4a090