🎬🖼️AGSoft Images To Video From Dir

🎬🖼️AGSoft Images To Video From Dir is the lazy version of the slideshow node, and I mean that as a compliment. Instead of wiring image inputs into the graph, you point it at a folder and it builds the video from whatever's inside. Same xfade transitions, same Ken Burns effect, same desync-proof voiceover - just with filesystem instead of wires.

This is the node to reach for when the images already exist as files: a batch of exported frames, a storyboard you generated elsewhere, or a folder of stills you want turned into a montage without dragging a hundred inputs onto the canvas. If you're re-rendering a project where the frames don't change, it's also much tidier than keeping the whole image pipeline alive.

How it works

folder_path points at the directory. The loader filters by extension via filter_type (*.*, *.png, *.jpg, *.jpeg, *.webp, *.bmp, *.gif, or custom), and custom_filter accepts multiple patterns - *.png, *.jpg or even a bare png jpg, which gets normalized to *.png. Files are naturally sorted, and you can narrow the selection with start_index / end_index (0-based, end_index 0 = to the end) and cap it with max_images (0 = all).

From there it's the same engine as the socket-based version: each slide shows image_duration seconds, joined by transition_type over transition_duration, optional ken_burns zoom/pan (kb_motion, kb_zoom_start/kb_zoom_end), and resize_mode/fit_mode/custom_width/custom_height for output sizing.

The voiceover has a nice split personality: audio_path takes a direct file path (priority), or you can connect the audio AUDIO socket instead. Either way the video renders first and the audio is aligned to its real length in a fast second pass (video copied, not re-encoded), so you don't get the classic trailing-silence or desync problems.

The inputs that matter

• folder_path + filter_type - where the images are and what counts as an image.
• start_index / end_index / max_images - grabbing a specific run of frames.
• Everything else is the slideshow controls you know from the sibling node: image_duration, transition_type, ken_burns, encoder_mode.

Outputs

The standard set: video_path (STRING), duration_seconds, duration_timecode, file_size_mb, width, height, fps, frames_est. Empty output_path → ComfyUI/output.

Install

Part of the comfyui-AGSoft pack. ComfyUI Manager → search "comfyui-AGSoft" → Install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/Art-xmaster/comfyui-AGSoft

Restart. No models, no heavy deps - it drives ComfyUI's bundled FFmpeg (or a system ffmpeg on PATH).

The one thing to double-check: start_index is 0-based and the sort is natural, so frame_10 comes after frame_9 (not after frame_1). If your slides come out in the wrong order, that's usually a naming scheme issue, not the node.