Wan2.2 FLF(First-Last Frame to Video) 14Bを試してみた
Wan2.2がFLF(First-Last Frame to Video)に対応したのを見つけたので、試してみます。
In case you don't know, with the latest ComfyUI updates by Comfy, we now have native support for WAN2.2 FLF2V. #comfyui pic.twitter.com/mSSmXAOpKL
— ComfyUI Wiki (@ComfyUIWiki) August 2, 2025
動画でも解説しています。
環境
OS:Windows 11
GPU:GeForce RTX 4090
CPU:i9-13900KF
memory:128G
環境構築
以下の記事を参考にして構築してください。
手順
ワークフロー
ワークフローは以下です。
解像度:480×832、全体フレーム:81フレーム

解像度:720×1280、全体フレーム:81フレーム

画像
開始画像

終了画像

プロンプト
Enchanting anime schoolgirl with black twin-tails speaks continuously while gracefully moving deeper into the refreshing river water. She starts by playfully adjusting her sailor collar, then slowly wades further into the stream with sensual, flowing movements. Her fitted uniform becomes increasingly wet and clings to her curves, creating beautiful silhouettes with each gentle motion. Natural breathing produces subtle chest movement while talking, wet hair streams beautifully around her face and shoulders. She maintains seductive eye contact with viewer while exploring deeper water, expression showing pure joy and cooling relief. Single continuous shot with smooth tracking. Sparkling water ripples, light refraction effects, and soft lighting enhance her feminine form and the translucent, clinging fabric.結果
解像度:480×832、全体フレーム:81フレーム
みはるが投稿
— kongo jun (@jun_kongo) August 2, 2025
Wan2.2 FLF(First-Last Frame to Video) 14B
・解像度:480×832
・全体フレーム:81
・生成時間:19分
・VRAM:22GB pic.twitter.com/UVuSIWNxgI
生成時間:19分
VRAM:22GB
got prompt
Using scaled fp8: fp8 matrix mult: False, scale input: False
model weight dtype torch.float8_e4m3fn, manual cast: torch.float16
model_type FLOW
unet missing: ['text_embedding.0.scale_weight', 'text_embedding.2.scale_weight', 'time_embedding.0.scale_weight', 'time_embedding.2.scale_weight', 'time_projection.1.scale_weight', 'head.head.scale_weight']
Requested to load WAN21
loaded completely 15478.626919891358 13627.902000427246 True
100%|██████████████████████████████████████████████████████████████████████████████████| 10/10 [08:30<00:00, 51.02s/it]
Requested to load WAN21
loaded completely 15478.626919891358 13627.902000427246 True
100%|██████████████████████████████████████████████████████████████████████████████████| 10/10 [08:37<00:00, 51.79s/it]
Requested to load WanVAE
loaded completely 3226.7562942504883 242.02829551696777 True
Prompt executed in 00:19:00解像度:720×1280、全体フレーム:81フレーム
みはるが投稿
— kongo jun (@jun_kongo) August 3, 2025
Wan2.2 FLF(First-Last Frame to Video) 14B
・解像度:720×1280
・全体フレーム:81
・生成時間:40分
・VRAM:23GB pic.twitter.com/Y2KDGWZ1IY
生成時間:40分
VRAM:23GB
got prompt
Using xformers attention in VAE
Using xformers attention in VAE
VAE load device: cuda:0, offload device: cpu, dtype: torch.bfloat16
Using scaled fp8: fp8 matrix mult: False, scale input: False
CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.float16
Requested to load WanTEModel
loaded completely 21474.8 6419.477203369141 True
C:\Users\XXXXX\comfy-ui\ComfyUI_windows_portable\ComfyUI\comfy\ldm\modules\attention.py:451: UserWarning: 1Torch was not compiled with flash attention. (Triggered internally at C:\actions-runner\_work\pytorch\pytorch\builder\windows\pytorch\aten\src\ATen\native\transformers\cuda\sdp_utils.cpp:555.)
out = torch.nn.functional.scaled_dot_product_attention(q, k, v, attn_mask=mask, dropout_p=0.0, is_causal=False)
Requested to load WanVAE
loaded completely 5235.522792816162 242.02829551696777 True
Using scaled fp8: fp8 matrix mult: False, scale input: False
model weight dtype torch.float8_e4m3fn, manual cast: torch.float16
model_type FLOW
unet missing: ['text_embedding.0.scale_weight', 'text_embedding.2.scale_weight', 'time_embedding.0.scale_weight', 'time_embedding.2.scale_weight', 'time_projection.1.scale_weight', 'head.head.scale_weight']
Requested to load WAN21
loaded partially 6710.994919891356 6706.5484046936035 0
100%|██████████████████████████████████████████████████████████████████████████████████| 10/10 [15:46<00:00, 94.62s/it]
Using scaled fp8: fp8 matrix mult: False, scale input: False
model weight dtype torch.float8_e4m3fn, manual cast: torch.float16
model_type FLOW
unet missing: ['text_embedding.0.scale_weight', 'text_embedding.2.scale_weight', 'time_embedding.0.scale_weight', 'time_embedding.2.scale_weight', 'time_projection.1.scale_weight', 'head.head.scale_weight']
Requested to load WAN21
loaded partially 6704.994919891356 6704.990783691406 0
100%|█████████████████████████████████████████████████████████████████████████████████| 10/10 [21:53<00:00, 131.37s/it]
Requested to load WanVAE
loaded completely 3178.1967163085938 242.02829551696777 True
Prompt executed in 00:40:16感想
light2xvとかを試せてないので、生成時間が遅い。
品質はそこそこですね。Wan2.1の時よりは優れていると思います。
(Wan2.1のI2Vほどの感動はないかな。)
