Skywork AI Skyreels V4 First Last Frame (Official Edition)

Text-to-video gives you a middle with no ending. If you've ever generated a clip and wished you could just say "and it ends here," first-last-frame (FLF) is the fix: you give Skyreels V4 a starting frame, an ending frame, and a prompt, and the model figures out a believable way to get from one to the other.

What this node does

BizyTRD_Skyreels_V4_Official_FLF2V_API generates a video between two anchor images. The classic use is match cuts: shot A of a character at a door, shot B of them across the room, and the node invents the transition. It's also the tool for looping-ish shots where you want the end state locked in, and for storyboards where you've already drawn the keyframes and need the in-betweens.

It's cloud execution, like every BizyTRD node - flip the BizyAirPlus button ON, and the generation happens on BizyAir's servers while ComfyUI waits and streams back progress. Your GPU never touches the model.

The inputs that matter

• first_frame_image - where the clip starts. Required.
• end_frame_image - where it ends. Also the interesting part: the tooltip notes formats JPG/PNG/GIF/BMP and says at least one of first/end/mid must be provided. There's also an images autogrow list if you want mid-point anchors on a longer run.
• prompt - the motion and feel between the anchors.
• resolution - 1080p by default.
• duration - 3 to 15 seconds.
• mode - std (default) balances speed and quality; fast is quicker but currently only works with sound off and doesn't support audio-video output yet.
• sound - on by default: V4 generates synchronized dialogue and sound effects, not just silent footage. Flip it off for a silent take.

Shared BizyAir extras apply: api_config for a per-request endpoint/key override, skip_error to degrade gracefully instead of aborting the whole graph on a cloud error.

Outputs

• video - the finished clip; wire into a preview or save node.
• request_id - cloud task ID for logs and support.
• urls - hosted URLs of the result.

How to install

cd /path/to/ComfyUI/custom_nodes
git clone https://github.com/siliconflow/BizyAirPlus.git
cd BizyAirPlus
python -m pip install -r requirements.txt

Restart (or grab "BizyAirPlus" from ComfyUI Manager). The deps are bizyair-cloudberry and bizytrd; nothing heavy downloads because nothing runs locally. First run: account at bizyair.ai → API key → enter it when the button prompts (Settings > BizyAirPlus > API Key works too; key lives at ~/.BizyAirPlus/apikey.ini).

Where people get burned

Two traps. The first is the fast mode: it looks like a free speedup, but it forces sound = false, so if you leave sound on and flip to fast you'll get a failure or a silent surprise. The second is anchor drift - if your first and last frames are wildly different subjects, the model has to invent a bridge and it shows. Keep the anchors visually related and let the prompt carry the action.

And if you hit the "is this malware?" wall: that's leftover from ComfyUI Manager auto-installing BizyAir for unrelated workflows back in 2025. It's a legit SiliconFlow pack. Accidental installs should be removed with pip uninstall bizyair-cloudberry bizytrd, not by deleting the folder alone.