comfyui-AGSoft

Glue TTS clips into one track with crossfades and matched loudness

Pull the audio out of a generated clip (or just part of it)

Swap a video's audio without re-encoding the video

Cut a chunk out of an mp3 without re-encoding it

Audio Split, but with fades, loudness matching, reverse, and looping

Face crops on autopilot — RetinaFace shipped inside the pack

AGSoft Crop Stitch

Paint your mask's shape onto the image, in any color

An empty latent with presets, megapixels, and sensible rounding

FLUX.2 latents without doing the ÷16 in your head

Krea 2's 16-channel latent, without the crash course

The 112 rule, automated — QwenImage latents that just work

A float constant you can round, and not a widget more

A proper before/after viewer, not a screenshot collage

Side-by-side composites without the manual resize first

Turn an image batch into a grid, without a single resize node

Crop with presets, percentages, and a mask that comes along

Actually draw your crop box in the ComfyUI graph

Resize image and mask together — proportions preserved, edges intact

The resize node that keeps your mask honest

Rotate an image, and take its mask along for the ride

Stitch 2–4 images into one canvas without the alignment guessing game

Turn a batch of frames into an actual slideshow with motion

Slideshow straight from a folder — no image wiring at all

Pad an image and get the outpainting mask for free

Padding with a target resolution, not just 'add 64 pixels'

The tiny resize node that does exactly what you ask, no more

Resize by megapixels, not by guesswork

Crop tight, inpaint big, stitch it back — the whole trick in one node

The stitch step that makes crop-and-stitch inpainting actually work

The world's simplest integer node — and why it's in the pack

A KSampler that hands back the receipt

Grab the last frame of a video without decoding the whole thing by hand

The AGSoft audio loader that always hands you a real file path

One node, image plus mask, zero conversion spaghetti

Resize image and mask together, or watch your inpaint drift

Load a whole folder of images into one node, sizes and alpha intact

A video loader that accepts a path, a file picker, or another loader's output

Feed a list of 10 floats into a loop without a converter node

Ten integer inputs, one list out, zero dropped values

Ten text boxes that become one list — even the empty strings

A float range generator that includes the end value, rounding and all

Collect up to 30 image inputs into one list for batch processing

An integer range that doesn't quietly drop the end value

Sweep 20 LoRA strengths in one queue

Five unique seeds out of one node — stop re-rolling by hand

Twenty samplers in one wire — A/B testing without the graph spaghetti

Gather a dozen prompts into one list without a daisy-chain of nodes

LTX video math you no longer have to do by hand

Soft edges on your inpaint mask — the difference between 'seamless' and 'rectangle'

Two masks, five ways to combine them — mask math without the headache

Erode, fill, feather and binarize in one node

Turn any solid-color region into a mask — poor man's chroma key

Math in your graph, written as actual math — not five primitive nodes

The parameter calculator that keeps MiniMax H3 samplers happy

A bell that rings in your browser when the queue is finally done

The Flux reference-image workflow in one node — encode, attach, and sample

A save node that stops overwriting your work — dated folders, real formats, no collisions

Preview a whole batch before you commit — AGSoft Save Image Plus

See what's actually flowing through your graph — numbers, text, whatever

A text debug node that shows the actual prompt, right on the node

Style presets without 40 text boxes of dead prompt text

One switch that routes images, text, latents, models — anything at all

The prompt joiner that doesn't leave empty commas behind

Join exactly two strings without the stray-comma disease

Four prompt parts, joined once — the mid-size workhorse

Eight prompt parts in one join — the maxed-out template builder

Flip between two prompts with one integer — the A/B tester's switch

Four prompts, one selector — branch without building a tree

Eight prompt presets behind one selector — a style menu for your workflow

The plain multiline text box ComfyUI should have shipped

Ten text transforms in one dropdown — and one of them hands you a number

Burn a caption onto your render without ever leaving the graph

Find-and-replace for prompts, with up to three swaps in one pass

One text in, fifty named outputs — the node that feeds your loops

Free translation with no API key, fifteen backends, and a fallback chain

One node to upscale with a model OR plain resize — pick your poison

Join clips instantly with stream copy — when they match

Concatenate without the 'everything must match' lecture

Grab the exact frames you need from any video — first, last, or an even sample

Hold a frame so the viewer actually sees it

Burn text and watermarks onto video, positioned by expression

Convert, mux audio, or just preview

Start slow, end fast — or the whole clip at double time

Cut a fragment out of a video, fast or frame-accurate

Retime, reverse, or boomerang a segment of your clip