Genshin TCG Style [Wan 14B]

Trigger Word: Genshin_TCG
Model: Wan 2.1 t2i 14B
All examples are generated with 1.0 LoRA strength and CFG=6
For inference used Kijai's workflows

For prompt, I recommend using the following structure:

"Genshin_TCG medium shot" + character description (appearance, pose, clothing) + key object (weapon/artifact) + background + dynamic elements. Special attention is paid to color contrasts (dark armor vs glowing accents) and mystical atmosphere (starry sky, magical particles).

If you want to add a gold frame like in the TGC cards, add this to the end of the prompt:

The frame is adorned with a golden border featuring intricate designs, including star-like emblems in each corner and delicate patterns along the edges, creating an elegant and polished visual effect.

The previous version for Wan 1.3B can be found here https://civitai.com/models/1728768/genshin-tcg-style-wan-13b

Training Details

It is much easier to train the 14B version than the 1.3B. The movements are smooth, there are almost no artifacts during generation. For the training, a dataset of 54 short videos with cards from the Genshin Genius Invocation TCG card game was used. Since I used diffusion pipe for training, I'll just post the toml files.

For dataset:

resolutions = [[514, 304]]
enable_ar_bucket = true
min_ar = 0.5
max_ar = 2.0
num_ar_buckets = 7
frame_buckets = [1, 32, 33]

[[directory]]
path = "/home/user/Genshin_TCG_dataset/videos/304_514"
num_repeats = 5
resolutions = [[514, 304]]

[[directory]]
path = "/home/user/Genshin_TCG_dataset/videos/368_620"
num_repeats = 5
resolutions = [[620, 368]]

[[directory]]
path = "/home/user/Genshin_TCG_dataset/videos/492_828"
num_repeats = 5
resolutions = [[828, 492]]

For train:

output_dir = "/home/user/Genshin_TCG/14B"
dataset = "/home/user/config/dataset/dataset_v001.toml"

epochs = 80
micro_batch_size_per_gpu = 1
pipeline_stages = 1
gradient_accumulation_steps = 1
gradient_clipping = 1
warmup_steps = 10
eval_every_n_epochs = 1
eval_before_first_step = true
eval_micro_batch_size_per_gpu = 1
eval_gradient_accumulation_steps = 1
save_every_n_epochs = 1
activation_checkpointing = 'unsloth'
partition_method = "parameters"
save_dtype = "bfloat16"
caching_batch_size = 1
steps_per_print = 10
video_clip_mode = "single_beginning"
blocks_to_swap = 32

[model]
type = "wan"
ckpt_path = "/home/user/Wan2.1-T2V-14B"
dtype = "bfloat16"
transformer_dtype = "float8"
timestep_sample_method = "logit_normal"

[adapter]
type = "lora"
rank = 64
dtype = "bfloat16"

[optimizer]
type = 'AdamW8bitKahan'
lr = 5e-5
betas = [0.9, 0.99]
weight_decay = 0.01
stabilize = false