You can use the online workflow below. If the results are good, you can then try setting it up and running it locally.
👉 Click Here For Online WorkFlow
As a fan benefit, you’ll receive 1,000 points upon registration, plus 100 points for daily logins. Enjoy a smooth experience with an RTX 4090 and 48GB of VRAM.
Model Description
This is an upgraded version of the previously released Kontext Figure Maker, trained on the Qwen Image Edit model, with results far, far better than the Kontext version!
This version can reliably produce realistic figures in a style similar to Nano Banana (now Gemini). It also allows you to change the original image’s scene by switching prompts—see the prompt recommendations for details. And don’t forget the excellent Chinese support in the Qwen Image series.
Recommended Settings
Sampler: Euler
LoRA Weight: 1
CFG: 1.0 (Using the Qwen-Image-Lightning-8steps-V1.1 , acceleration LoRA weight 1.0)
Steps: 8
Recommended to use high-resolution upscale: 1.5× scale. Denoising set to 0.5. Beta scheduler is recommended, with 8 steps, CFG 1.0, using the Qwen-Image-Lightning-8steps-V1.1, acceleration LoRA weight 0.5.
Recommended Prompt (In Chinese)
You can also write your prompts in English. Most of the sentences and words in the prompt can be replaced, so feel free to explore!
Trigger Word：补全主体的躯体，保持姿势和面部特征不变，将主体变成一个写实的3d pvc手办
Other Prompts：
补全主体的躯体，保持姿势和面部特征不变，将主体变成一个写实的3d pvc手办，在手办的后面，有一个比手办大的设计精美的二次元手办包装盒，包装盒上写着精美设计的“千问QWEN”艺术字，包装右下角有正方形的红色标签，红色标签上写着“ALDNIKI”，“ALD”和“NIKI”各占标签的一行，包装盒上印着图片形象,在包装盒后面还有一个AOC牌子的电脑屏幕，屏幕上显示着3d建模软件中手办的灰色3d模型，手办被摆放在一个半透明的展示盘上，强调手办的pvc塑料质感，场景在一个手办生产工作室内，后面有3d打印机
补全主体的躯体，保持姿势和面部特征不变，将主体变成一个写实的3d pvc手办，手办被拿在一只真实的手里把玩，强调手办的pvc塑料质感，场景在一个手办生产工作室内
补全主体的躯体，保持姿势和面部特征不变，将主体变成一个写实的3d pvc手办，手办被放置在漫展摊位上进行售卖，后面有正在售卖的工作人员的身体
补全主体的躯体，保持姿势和面部特征不变，将主体变成一个写实的3d pvc手办，手办置于会发光的魔法法阵中央，法阵浮现半透明光线和神秘符文，手办表面带有逼真PVC光泽，法阵散发蓝色和金色光芒在手办上投射动态光影，背景是古老手办工作室与奇幻元素结合的场景，桌面上散落着模型工具和魔法书籍，整体画面呈现幻想实验室的神秘氛围。
补全主体的躯体，保持姿势和面部特征不变，将主体变成一个写实的3d pvc手办，手办被置于未来感浮空展示架上，周围漂浮着微型全息屏幕显示手办设计信息，光线从架顶射下形成逼真阴影，PVC表面细节清晰，背景是高科技手办展览厅，氛围冷色调科技感十足，同时有微型机械臂在旁辅助整理手办道具。
(Use your imagination)
模型介绍
这是此前发布的Kontext Figure Maker的升级版本，基于Qwen Image Edit模型训练，效果比Kontext版好的多得多得多！
此版本可以稳定地产出类似Nano Banana (now Gemini) 风格的真实手办，同时此版本可以尝试切换提示词来更改原图的场景，详见提示词推荐。同时不要忘了Qwen Image系列的中文支持。
推荐设置
采样器: Euler
LoRA 权重: 1
CFG: 1.0 （使用 Qwen-Image-Lightning-8steps-V1.1 , 加速LoRA权重1.0)
步数: 8
推荐使用高清放大: 1.5倍. 降噪 0.5. Beta 调度器, 8步, CFG 1.0, 使用 Qwen-Image-Lightning-8steps-V1.1, 加速LoRA权重0.5
推荐提示词
你同样可以使用英文来编写提示词, 提示词中的大部分句子和词语都可以被替换，请尽情探索！
触发词：补全主体的躯体，保持姿势和面部特征不变，将主体变成一个写实的3d pvc手办
其他提示词：
补全主体的躯体，保持姿势和面部特征不变，将主体变成一个写实的3d pvc手办，在手办的后面，有一个比手办大的设计精美的二次元手办包装盒，包装盒上写着精美设计的“千问QWEN”艺术字，包装右下角有正方形的红色标签，红色标签上写着“ALDNIKI”，“ALD”和“NIKI”各占标签的一行，包装盒上印着图片形象,在包装盒后面还有一个AOC牌子的电脑屏幕，屏幕上显示着3d建模软件中手办的灰色3d模型，手办被摆放在一个半透明的展示盘上，强调手办的pvc塑料质感，场景在一个手办生产工作室内，后面有3d打印机
补全主体的躯体，保持姿势和面部特征不变，将主体变成一个写实的3d pvc手办，手办被拿在一只真实的手里把玩，强调手办的pvc塑料质感，场景在一个手办生产工作室内
补全主体的躯体，保持姿势和面部特征不变，将主体变成一个写实的3d pvc手办，手办被放置在漫展摊位上进行售卖，后面有正在售卖的工作人员的身体
补全主体的躯体，保持姿势和面部特征不变，将主体变成一个写实的3d pvc手办，手办置于会发光的魔法法阵中央，法阵浮现半透明光线和神秘符文，手办表面带有逼真PVC光泽，法阵散发蓝色和金色光芒在手办上投射动态光影，背景是古老手办工作室与奇幻元素结合的场景，桌面上散落着模型工具和魔法书籍，整体画面呈现幻想实验室的神秘氛围。
补全主体的躯体，保持姿势和面部特征不变，将主体变成一个写实的3d pvc手办，手办被置于未来感浮空展示架上，周围漂浮着微型全息屏幕显示手办设计信息，光线从架顶射下形成逼真阴影，PVC表面细节清晰，背景是高科技手办展览厅，氛围冷色调科技感十足，同时有微型机械臂在旁辅助整理手办道具。
(发挥你的想象)
Description
FAQ
Comments (40)
Hey, first of all, nice work on the LoRA. I've been wondering if you have done any experience with a lower/halved rank/network dim size to create a smaller file because ~560mb is quite hefty for most concepts I see being made on CivitAI for Qwen image, while LoRAs of half the size seem to do the job too. I would appreciate it if you could train halved rank LoRAs in the future if the quality is no different or negligible with a halved rank compared to your current setting and the concept isn't completely unknown to the model (like this one which base Qwen Edit do fairly decently already)
Thanks a lot for the detailed feedback and suggestion! You’re absolutely right — the current rank/network setting makes the file pretty large. I haven’t experimented much yet with lower-rank LoRAs for Qwen Edit, but I agree it’s worth testing, especially since the base model already handles this concept decently. I’ll definitely try training a halved-rank version in the future and compare the quality. If the results are comparable, I’ll release a smaller “lite” version alongside the full one. Really appreciate your input!
Nice, this looks amazing!
Thanks :D
Woha! those are awesome results! all my training attempts with ai toolkit looked meh. curious to ask - what trainer did you use to get such results? could you share more details of your training? how much steps/dataset size/other details that could help achieve such results?
I’m using aitoolkit with a dataset of around 40 images. Since it’s a small dataset, the images must be carefully selected, and the annotations must be precise. For a dataset of this size, about 6000 training steps are sufficient. For my other LoRAs with larger datasets, I usually increase the number of training steps.
This is awesome! I'm using it to make figurines from realistic images, and about one in every 10 or so generations does the facial likeness match significantly. Any tips on getting it more consistent to copy facial likeness?
Thanks for the feedback! There aren’t that many real-person samples in the training dataset, so using actual photos does take a bit of luck :P
THIS IS AMAZING, perfect fit using Qwen! Thanks — way better than Nano Banana.
Thank you :D
如果图片是背后视角的，脚掌都成正面的，有没有办法弄到正确啊？
那就用背面全身照试试
😍 👍🏻
I'm surprised to see there's a lora for this concept, I did that with normal Qwen by just prompting... Oh well, let's try and see the result
That’s right — the Qwen edit base model itself can achieve the figure conversion effect, but the results are rather unsatisfactory, especially when dealing with half-body shots, which turn out quite poorly.
Can someone please help me and tell how to properly use this? I never use Comfy and trying to turn photo to figure, but all i got - blurry edited image. I download some workflows from post. And all i got - errors or blurry image.
I have rtx 3060 laptop with 6GB VRAM.
Just use the comfyui default Qwen Image Edit workflow. Add in lora. Paste in the trigger words and Prompt into the Prompt box. In english or Chinese. Are you using Image-Edit?
@mattinglyjohn722 thanks, i double checked and i was used qwen-image and not image-edit. Now everynthin working!!
6g 都可以跑image-edit吗
@344249382151 我11G显存直接跑GGUF的Q8版本。6G应该也是可以的，只要不启用共享显存速度就不会慢得太离谱，很多人有着不错的配置，用起来慢很大可能就是设置不当，COMFYUI直接把显存占满然后走共享显存了。
HI there, similar issue, I'm not getting this to work at all. I've tried with qwen image Q8 and Qwen Imge Edit Q8 and neither work. My end image is just the same as my input image, but kinda messed up. Any ideas?
@Delavestra 输出和输入一样的话也许是工作流的线路连接问题，也许是降噪<1（小于一）
This is so cool! What would you say is superior, the Qwen Edit or Flux Kontext version of your lora?
Nice model! 可以问一下是用什么工具训的LoRA嘛
老大，似乎Qwen/Qwen-Image-Edit-2509就用不了这个lora了，但是Qwen/Qwen-Image-Edit-2509增强了很多性能，求大佬更新
It technically works with it (using the official ComfyUI workflow for 2509 as a base) but there is some more or less noticeable moire effect like pattern on the result images.
‘我试了好几个Lora，还用过一个动漫替换成真人的，跟你一样用Qwen-Image-Edit-2509Lora失灵，几次测试后发现只要修改环境位置或修改动作就能生效，必须换场景。
Two remarks after using this LoRA for a few hundred images:
1. it really likes to add transparent support rods behind the figures, even when it is absolutely not needed, for eg. figure firmly standing on two feet. It can't really be avoided with negative prompting as the LoRA doesn't seem to have a concept of the support rods. I think this should be more controllable. (Of course the rod can be removed with a second Qwen Image edit run without the LoRA, and specifically asking to remove the rod, but it would be better if this wouldn't be needed)
2. the LoRA likes to create big breasts even for more moderately chested girls. This isn't a problem most of the time when clothing covers the chest, but when there is even the slightest hint to the actual shape of the breasts, then the LoRA likes to jump on the opportunity, and make the breasts quite prominent and bigger, and in some cases the result is very bad looking. Prompting doesn't have any effect in this case either. This also should be more controllable. Was the LoRA maybe trained only with big chested figures?
Why not put in english first? (Doesnt work at all - Time losted!)
I don't speak Chinese at all and I don't speak English fluently but I managed to make it work.
Just copy/paste the trigger words (in Chinese) in your prompt, then add your English text into that prompt (for ex, ",put the figurine on a glass table in the living room")
it works very well, even with French Belgian comics styles.
does this lora produce images that appear as figures or does it produce a 3d file somehow and the images below are real 3d printed figurines? I appreciate the help
it produce images that appear as figures
wow i can do that brief trend with nsfw things, thanks lol
This LoRA can create unique styles! Best a2r!Awesome!
hello, why does this NOT work with qwen models, 2511, 2509, but works with klein 9b model? bizarre, but that is what is happening with the wf, https://www.runninghub.ai/post/1961347151148290049/?inviteCode=rh-v1013
trying a different qwen wf, but this tells me, the prompt is what is working?
Hi, I made fantastic outputs when I tried this setup weeks ago, and now suddenly I got this message (even after updating all): (and used already made image to start workflow, yours btw)
LoraLoaderModelOnly: - Return type mismatch between linked nodes: model, received_type(WANVIDEOMODEL) mismatch input_type(MODEL)
Can you help?
This doesnt seem to work at all. really sad
Details
Files
aldniki_qwen_figure_maker_v01.safetensors
Mirrors
aldniki_qwen_figure_maker_v01.safetensors
aldniki_qwen_figure_maker_v01.safetensors
aldniki_qwen_figure_maker_v01.safetensors
QWEN_FigureMaker.safetensors
aldniki_qwen_figure_maker_v01.safetensors
aldniki_qwen_figure_maker_v01.safetensors
aldniki_qwen_figure_maker_v01.safetensors
aldniki_qwen_figure_maker_v01.safetensors
aldniki_qwen_figure_maker_v01.safetensors
aldniki_qwen_figure_maker_v01.safetensors
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.