ComfyUI Timesaver Nodes
ComfyUI nodes from Timesaver.
Nodes (88)
Point the camera, get the prompt for that exact view
Actually see your animation before you pay for the encode
The audio loader with a waveform you can actually use
Audition audio in the graph before you commit to a file
SoundCloud-style waveform images, generated in the graph
Stop doing tile math in your head
The name lies — it loads one image, and that's the point
Run a hundred prompts through the same workflow
Turn one graph into a hundred independent runs
Get your results on disk before the batch finishes
BiRefNet background removal with hair you can believe
Make every shot agree with one reference frame
The before/after wipe that doesn't lie about video
Crop to the interesting part, process only that
This node borrows your GPU's game upscaler for real frames
Your Qwen Edit Came Back Three Pixels Off — Here's How to Put It Back
Download every model a workflow needs, in one shot
Cinematic film looks without leaving the graph
A clean float knob, and not a math node in sight
Buttery-smooth video from choppy model output
A wire that clears your GPU at the exact right moment
How big is that image, actually?
The longer side or the shorter side — pick one
A bypass switch for every group in your workflow
Make MiniMax H3 say the line you actually recorded
Design Ideogram 4 prompts visually, not in JSON
Trim the junk frames off the front and back
Split and iterate
Depth maps for stills that actually hold their range
The glue node that turns a list of images back into a batch
Get an LLM's prompt into the sampler without rewiring anything
The resize node that finally covers every mode you actually use
Stitch processed tiles back into one seamless image
Cut a huge image into overlapping tiles your GPU can actually handle
A clean integer knob so you don't drag a math node onto your graph
Krea 2 Keeps Ignoring You — This Is the One Number That Fixes It
Paint over the tourist, watch the watermark vanish, no Photoshop trip required
An inpaint sampler that makes the repair agree with its surroundings
The right way to upscale MiniMax H3 video without melting your GPU
A Checkpoint Loader for a File Nothing Else on Your Disk Can Open
The `UNETLoader` for Locked Files, Plus the Detail Everyone Else Would Have Missed
For Video Graphs There's No CLIP Wire — This Is the LoRA Loader That Admits It
Six LoRAs in one node, reorder by dragging, and it expands into the real thing
Your turbo LoRA is getting rounded off before it ever runs
Pick which LTX stage wins before you decode, so the loser never runs
Pin the start and end frames, let LTX interpolate between them
The one decode that gives you both a watchable preview and a true HDR master
One node per guide frame that feeds both LTX stages — and never runs the wrong branch
Read an EXR the way the HDR pipeline needs — without flattening everything above 1.0
One switch that turns the whole LTX 2.5 HDR path on — and off for free
Is the HDR range still there? This node answers with numbers
The same VAE file, forced to float32 — because that's where HDR lives
Two integers in, one integer out — the math node that stops the Primitive tangles
From a sloppy SAM mask to real hair-level transparency
Stop Googling \"832×480 or 1280×720?\" — Pick the Model and Let the Node Do the Maths
Three reference images into the conditioning stream, one node
Karaoke, remixes, and clean vocals — all inside the graph
At CFG 1 Your Negative Prompt Is Decoration — NAG Puts It Back
Wildcard prompts that don't put a swimsuit on a winter street
51 edit prompts someone already tested, free to copy
A local Qwen VLM that captions, describes, and writes your prompts
The 'always a valid canvas size' node you'll stick in every workflow
Paste the sharpened face back without a visible seam
Gemma 4 sitting in your graph, no buttons, no API key
Real-time upscaling on NVIDIA's hardware — speed over invention
Stop hand-typing SAM3 point JSON — just click the picture
The sampling schedule your Qwen turbo LoRA was actually distilled on
The Russian text fixer that stops TTS from saying the wrong word
Russian text-to-speech that lives inside your graph
Batch two images without breaking when one isn't there
The one-node inpaint that does the crop, sample, and stitch itself
A switch that doesn't blow up when a branch isn't wired
The lyrics box that puts the stress where you meant it
157 art styles you can browse instead of guessing the phrase
Speak your idea, get a production-ready prompt back
Prompt enhancement on Google's phone runtime — twice as fast, a fifth of the VRAM
Inpaint any model — no Fill checkpoint, no ControlNet, just this sampler
Your Qwen decode has a 2-pixel grid — TS VAE Fix deletes it before your upscaler eats it
The trim node that keeps picture and sound in sync
Steady depth maps for whole clips — not 24 flickering stills
Turn video_info into numbers you can actually route on
A video loader with a timeline, so a two-second cut doesn't cost an hour
H.264 to share, ProRes to edit, EXR to master
Local transcription with SRT out — no API key, no cloud
Your Resolve markers, turned into a YouTube chapter list in one step
Timesaver Nodes for ComfyUI
Nodes for images, video, audio and prompts, plus tools for organizing workflows.
Русский · Workflow examples · License
</div>Installation
In ComfyUI Manager, search for Timesaver. For a manual installation, run these commands from the ComfyUI directory:
cd custom_nodes
git clone https://github.com/AlexYez/comfyui-timesaver
cd comfyui-timesaver
python -m pip install -r requirements.txt
Use the Python interpreter that runs ComfyUI. Restart ComfyUI, then search for TS on the canvas. No separate frontend build is required.
To update a manual installation, run git pull --ff-only from the pack directory, install updated requirements with the same interpreter and restart ComfyUI.
<a id="-node-reference"></a> <a id="image"></a><a id="video"></a><a id="hdr"></a><a id="audio"></a><a id="llm"></a><a id="text"></a><a id="ideogram"></a><a id="files"></a><a id="loaders"></a><a id="utils"></a><a id="conditioning"></a>
Using the nodes
- Image — resize and alignment, masks, inpainting and restoration.
- Video — loading, trimming, saving, upscaling and HDR tools.
- Audio and prompts — audio editing, speech tools, prompt construction and model-assisted rewriting.
- Workflow tools — sliders, switches, file downloads and graph layout commands.
Open a node’s help in ComfyUI for its inputs and options. The same English and Russian reference and example workflows are available here.
Some nodes need model weights, an optional Python package or a separately installed application. Check the node's input tooltips and follow any dependency message in the ComfyUI console. If a node is missing, look for [TS Timesaver] in that console. Start ComfyUI with TS_VERBOSE_STARTUP=1 for the full registration report. See package metadata for version and dependency details.
На русском
Timesaver добавляет ноды обработки изображений, видео, звука и промптов, а также инструменты для работы с графом.
Подсказки по конкретным нодам доступны в ComfyUI и в справочнике. Готовые графы собраны в примерах workflow.
В ComfyUI Manager найдите Timesaver. Для ручной установки клонируйте пак в custom_nodes и установите зависимости командами выше. Используйте Python, которым запускается ComfyUI. После перезапуска найдите ноды по TS в поиске канваса.
Для ручного обновления выполните git pull --ff-only в папке пака, установите обновлённые зависимости тем же Python и перезапустите ComfyUI. Для отдельных нод могут понадобиться модели, дополнительные пакеты или внешние приложения. Смотрите подсказки к входам и сообщения в консоли ComfyUI. Если нода не появилась, найдите в консоли [TS Timesaver]. Флаг окружения TS_VERBOSE_STARTUP=1 включает полный отчёт регистрации при запуске.
Код распространяется по лицензии MIT. У сторонних моделей, SDK и приложений собственные условия.