DD Image To Video

Here's the thing about image-to-video models: most of them want a sequence of frames as input, not a single image, and their wrappers can be picky about batch shape. DD Image To Video takes one still image and repeats it into a frame sequence of a length you specify - total frames, actual duration, and fps all handed back as outputs. It's the "make this one frame look like a video so the video model will talk to me" node.

How it works

The mechanism is embarrassingly simple and that's the point: total_frames = round(duration × fps), then the input image tensor is repeated that many times into a batch. If you ask for 5 seconds at 30 fps, you get a 150-frame batch of the identical frame. It generates in chunks of 批处理大小 (Batch Size, default 30) so memory stays flat even for long sequences - you're never holding a 3000-frame tensor at once.

Inputs are 图片 (Image), 时长 (Duration in seconds, 0.1–300), 帧率 (FPS, 1–120), and that batch size. Outputs are the 视频帧 (Video Frames) IMAGE batch, plus three metadata values that make it genuinely useful: 总帧数 (Total Frames, INT), 实际时长 (Actual Duration, FLOAT - note it reflects the rounding, so 5s at 30fps comes back as 5.0 but odd ratios give you a slightly different number), and 帧率 (FPS, FLOAT).

When you'd reach for it

Feeding I2V models is the main event. Wan's image-to-video path, and several other video samplers, want a leading image as the first frame of a clip; this node gives you a placeholder sequence when you're testing a workflow before the real motion happens, or when a wrapper needs a full-length batch even though you only have a still. It's also useful for building test clips, checking that your video-combine/save pipeline handles the shape, and generating a "hold on this frame" segment. The honest read: it will never make a video move - that's the video model's job - but it removes the shape-mismatch failure before that job even starts.

Installation

Part of the DD-Nodes pack:

cd ComfyUI/custom_nodes
git clone https://github.com/Dontdrunk/ComfyUI-DD-Nodes
cd ComfyUI-DD-Nodes && pip install -r requirements.txt

Restart ComfyUI or use ComfyUI Manager (search "ComfyUI-DD-Nodes"). Pure torch tensor work - no extra models.

Where people get burned

Don't reach for this to animate - there are zero motion parameters because there's no motion to control; it's a still held for N frames. The 实际时长 rounding means "5 seconds" can come back as 4.97s or similar on awkward fps choices, which matters if you're matching audio or another clip's length. And the output is a batch of identical frames, so any downstream node that assumes temporal variation will do nothing interesting - again, expected. As with the whole pack, labels are Chinese-first with an English locale available.