Stack_WanImageToVideo

Of all the Wan video nodes in Apt_Preset, this is the one you'll reach for most. Everything else in the pack's Wan lineup - Fun Control, Animate, camera control, Phantom, HuMo, S2V - is a specialized variant. Stack_WanImageToVideo is the plain I2V mode: give it a picture, get back motion. It's also, not coincidentally, the community's favorite way to use Wan generally - anchoring the first frame to a known image sidesteps a lot of the consistency problems that plague pure text-to-video, which is why I2V became the default workflow shortly after Wan launched.

How it works. start_image is your source picture. clip_img and clip_vision_name (default clip_vision_h.safetensors) handle CLIP vision conditioning of that image - the standard mechanism every I2V variant in this pack shares. width/height default to 832×480, the classic Wan 480p footprint. Worth noting: length defaults to just 40 frames here, shorter than most of this pack's other Wan nodes (which mostly default to 77-81) - that's roughly 2.5 seconds at 16fps rather than the community-standard ~5 second clip, so if you're chasing the usual length, bump it up toward 81 rather than assuming the default is already tuned for a full clip.

Settings worth knowing beyond what's on this node. Community practice for Wan I2V: 20-30 steps and CFG 3.5+ on the high-noise pass if you're running without a speed-distillation LoRA; if you are using one (lightx2v, CausVid), CFG needs to drop to 1.0 with just 3-4 steps per pass, and the best-regarded compromise is applying the speed LoRA only to the low-noise pass while keeping the high-noise pass clean - that combination is what people credit with getting speed-LoRA quality much closer to full-quality Wan 2.2. Prompting for I2V specifically wants two genuinely separate descriptions, not one: describe the still image flatly and technically as if captioning a training dataset, then write a completely separate motion-only description for the video that doesn't restate what's already visible in the image - that split is the single most-repeated piece of Wan prompting advice in the community.

Installing it. No extra dependency for this node itself: git clone https://github.com/cardenluo/ComfyUI-Apt_Preset into custom_nodes, restart ComfyUI, or search the pack title in ComfyUI Manager. You need a Wan I2V checkpoint (2.1 or 2.2, Apache 2.0 and freely downloadable from Alibaba's official HuggingFace releases) and the CLIP vision model matching clip_vision_name in models/clip_vision.

Common issues. Videos that barely move or look like a slideshow are almost always a prompting problem, not a settings one - check you're not accidentally repeating your still-image description in the motion prompt, since Wan tends to interpret restated static description as "stay still." Blurry, plastic-looking skin and flattened lighting after adding a speed LoRA is the well-documented tradeoff of running distillation LoRAs at full strength on both passes - move it to the low-noise pass only if quality matters more than speed. And going past 81 frames without VACE or a chunking extension will reliably start producing looping or degrading motion - that's a hard architectural ceiling on native Wan generation, not a bug you can tune around with sampler settings.