Mickmumpitz
patreon
[NEW VIDEO] Create HYPERREALISTIC Consistent AI Characters - FREE & LOCAL!
🕑 Added 2025-10-07 15:55:47 +0000 UTCCreating hyperrealistic consistent characters is now easier than ever - and it's completely FREE!
All you need is one input image in any style. With a single click, my workflow automatically generates a perfect dataset: high-resolution upscaled images in different settings and poses, complete with detailed captions. If you want, you can even load in pose references or dress your character using the virtual try-on section of the workflow.
All of this is based entirely on free, open-source models that you can run on your own computer. But I'm also going to compare the outputs with commercial models so you can decide which approach works best for you.
By the end of this tutorial, you'll know how to create perfect humans or stylized characters, how to train your own LoRA model of these characters if necessary, and how to generate extremely realistic and perfectly consistent HD videos of your characters—all locally on your own computer.
- FREE GUIDE here: https://mickmumpitz.ai/guides/consistent-character-creator
- Runpod [Affiliate]: https://runpod.io?ref=o2xn5klc
- ComfyUI Download: https://docs.comfy.org/installation/comfyui_portable_windows
WORKFLOWS
🔗Download the FREE WORKLFOWS below. 👇
🔗 You can find the FREE QWEN IMAGE EDIT workflow here: https://www.patreon.com/posts/introducing-my-139998677?utm_medium=clipboard_copy&utm_source=copyLink&utm_campaign=postshare_creator&utm_content=join_link
🔗ADVANCED versions and exclusive example files, loras & datasets here: https://www.patreon.com/posts/advanced-version-140672384?utm_medium=clipboard_copy&utm_source=copyLink&utm_campaign=postshare_creator&utm_content=join_link
What You'll Learn:
I'll walk you through my complete workflow for creating hyperrealistic consistent characters - all you need is one single input image. From there, you'll discover how to automatically generate perfect training datasets with detailed captions in one click.
You'll learn how to train your own custom character LoRAs for maximum quality and consistency, then master my "Consistent Character Creator" workflow that uses Qwen Image Edit to transfer poses, swap outfits with virtual try-on, and generate characters in any scene you can imagine.
Once your character is ready, I'll show you how to generate absolutely stunning hyperrealistic images using Wan 2.2, and then take it even further by creating cinematic HD and even 4K videos of your character in motion.
And because I know everyone's setup is different, I'll compare results between free open-source models and commercial options like NanoBanana and Seedream 4.0, so you can choose the approach that works best for your hardware and project needs.
251119 - UPDATE 3.7
Unfortunately, due to a ComfyUI update, I had to replace the image saver node, so the data set is no longer automatically saved in the correct folder. Instead, the images end up in the ComfyUI output folder. If you want to change this, you can manually enter an absolute path where your final data set will be saved.
![[NEW VIDEO] Create HYPERREALISTIC Consistent AI Characters - FREE & LOCAL!](https://img5.patrefans.com/storage/3/ev/hs/563145-019e8569-da13-7aca-8338-203b09f4d830.jpg)
Comments
Roman
Guys, I'm tired of problem with: VHS_LoadImagesPath directory is not valid I've read all the replies above but id didn't work, anyone solve this?
Vojta
I have a problem guys https://youtu.be/PhiPASFYBmk?t=620, here you click on dataset creation button but I dont have any there, what should I do? Please help!
Artokun
will we see a CCC_3-8 that swaps out flux1 with flux2?
Y
Same here, could really use a fix. AI is no help.
selcuk artut
I am getting VHS_LoadImagesPath error. It says the directory is not valid. Join String Multi and ImageScaleToMaxDimen are highlighted in red color. On another note; manually adjusting the output folder in the Save Image KJ didn't seem to help either
dexboygenius
I got it to work. I hoping to learn a little more. Like how to use loras and checkpoints
Rene Kolo
depends what you trying to do, but usually just comfyUI:latest and then ostris if you use that for training datasets
sandra andrade
How much do I have to pay to have access to the workflows? Because I´ve already paid the subscription
Alex Strothers
Ive paid a few days ago hoping to get a lora done with you detailed video.... I HAD NOTHING BUT PROBLEMS, IT WAS HARD TO FIND THE WORKFLOW, I AND NOT ALL THE NODES ARE AVAILABLE ANYMORE TO CREATE A LORE WITH YOUR WORK FLOW. COMPLETELY IRRITATING!!!!!
Brian Gebel
While running B - DATASET CREATION I noticed that when it resizes and upscale some of the images stretch. The primary input and turn around are good examples. I'm not familiar with training so not sure if this is intentional or not but I assume it is not?
Deborah Lees
Hi, I joined as a member but I've just spent the last hour trying to get onto your discord with no luck. Patreon & Discord are connected but your server is not showing up, I tried all the troubleshooting but can't figure it out so very frustrating. I've cancelled my membership but if you can help me connect so I can get some help to get started I'll try again.
MR
Which runpod template do I use?
MR
Hi Mick, Does this run on Mac?Please say yes, I'm tired of my Mac only giving problems with new technologies.
Chris Russell
Finally got this to work but to be fair I have been working nights and time spent has only been 30 mins time spent so few questions. The expressions the teeth were all fucked up and as someone who just broke a tooth it gives me ptsd vibes! Lol but all joking aside any recommendations on fixing the tooth’s?
DeafPerv
Can I input an image that has the full body with both frontal and rear view of the character? My characters are very specific and I don't want to rely on comfyui generate their body proportions.
Stevie Tee
Hi Mick, I'm trying your 251007_MICKMUMPITZ_CCC_3-6.json workflow, I've installed everything correctly but all I'm getting is black images, is there something I'm missing?
Robert Jacobs
yes, i have the same issue, it seems to me, that the difference is the input prompt, sometimes there is stringconstantmultiline and sometimes TextencodeQwenimageeditplus, which seems to be superior.
Robert Jacobs
Edit: does someone have the stock_photography_wan22_LOW_v1.safetensors file ? / ok so i found the wan22 version i hope : https://prompthero.com/ai-models/stock-photography-wan-2-2-download
Tom Gottsauner
gibts den advanced für character noch gar nicht ?
Tyler Durden
I think main thing that's needed is a way to turn off a bunch of these example groups, as usually you just want to work on one image, and it takes a really long time to run all this. If bypass the groups or the nodes in them, it breaks the workflow.
Tyler Durden
Same
Roman Protsyk
Hi Mick, I want to thank you for creating such a detailed and comprehensive flow for the character. The work you’ve done is really impressive, and I truly appreciate it. I have a few questions I’d like to get your advice on, and I’ll send the examples separately. Question 1: Some blocks, such as Back, Profile, and Close-up, generate results that are completely different from what’s specified in the prompt. Do you have any idea what might be causing this? All these blocks are the same; the only difference is the prompt. Could you also let me know which versions of Comfy, ComfyUI frontend, and Python you used for this flow? I’ve noticed that different builds sometimes behave differently. I noticed that some seed values differ across blocks, while in most blocks the seed is repeated since it’s fixed. Was it intended to be this way?
Roman Protsyk
I have the same problem. I tried both the advanced and the free versions of this flow, but the errors are the same. In some nodes (Concept ART, Scene 06, T-pose, Model Pose, and others), the model doesn’t follow the prompt and generates something completely different. The question is — how can this be fixed? I’ll also write to the author separately in a private message.
Åke
I managed to make it work somehow, but all the nodes. I get issues with the CLOSE UP, PROFILE and BACK. THey all generate the same image it doesn't matter what I do, even if I change the prompt. It doesn't generate the T-POSE, MODEL POSE, SIDE VIEW, BACK VIEW and WALKING. It says: "Prompt outputs failed validation: Required input is missing: images SaveImage:- Required input is missing: images". Since this images are not generated, the other ones depending on them neither get generated (POSE TRANSFER and VIRTUAL TRY ON). Any idea why this is happening?
BERT
Hi Mick. Where could I download the standalone uso flux upscaler model ? I know it is in your main CCC json but it is instanced and the nodes are hidden. I would need something more user friendly and the url is cropped on your video. Would appreciate this a lot. thank you in advance.
Ghazi N3li
Hi, i am facing this error in the node VHS_LoadImagesPath directory is not valid: ComfyUI/output/lunaheart7 anyone knows how to fix this
Simon Hope
Im getting this error "Error while deserializing header: header too large" related to the Lora loader when using https://huggingface.co/lightx2v/Qwen-Image-Lightning/blob/main/Qwen-Image-Lightning-4steps-V1.0-bf16.safetensors - any ideas?
Adam Smith
I haven't had much luck with the previous workflows as far as consistency goes, but I'll keep on trying!
Kevin Wilkinson
Lol , so it is rendering me out some images now and they are pretty funny. My good looking girl portrait has now got a huge belly pepper nose and looks as plastic as barbie lol. The emotions make her look like she's having a fit. Lol. I will see all the images in morning as im only on a slow PC
M Razor
podrias hacer una guia rapida de configuracion para ostris, entrenando diferentes modelos principalmente, wan, qween y flux, con el seteo complate desde las imagenes necesarias para el entrenamiento de cada modelo y la configuracion segun el modelo
