A specialized LoRA for Krea 2 Turbo, designed to generate structured subject reference sheets.
This LoRA focuses on the layout, pose, and reference-sheet structure rather than one specific person or character. It is made to help create sheets with full-body turnaround views and portrait/head-angle studies in a clean, organized format.
It can be used for realistic people, anime characters, fashion concepts, cosplay concepts, fantasy characters, male subjects, female subjects, stylized designs, and more.
Best results so far:
Aspect Ratio: 1:1 > 4:3
Resolution: 2 MegapixelsActivation Tag
Use this tag at the very beginning of your prompt:
subjectposesheet01Standard Activation Sentence
Recommended starting version:
subjectposesheet01. This image is a structured subject reference sheet showing the same subject multiple times in a consistent layout. The top row presents three full-body turnaround views consisting of a front view, a side view, and a back view. In each full-body view, the subject stands in a relaxed A-pose with both arms extended outward. The lower section contains multiple portrait head-angle studies, including upward-looking angles, neutral front-facing views, side and three-quarter views, and downward-looking angles. A larger front-facing portrait appears on the lower right.After this sentence, describe your own subject.
Stronger Activation Sentence
Use this version if you want to push the pose and sheet layout more strongly:
subjectposesheet01. This image is a structured subject reference sheet showing the same subject multiple times in a consistent layout. The top row presents three full-body turnaround views consisting of a front view, a side view, and a back view. In the front view, the subject faces the camera directly, standing upright with the feet about shoulder-width apart, the torso straight, both arms extended outward in a relaxed A-pose, elbows mostly straight, hands open, and palms slightly angled forward. In the side view, the subject is shown from the profile while keeping the same upright A-pose with one arm extending toward the viewer and the other arm extending away from the body. In the back view, the subject is shown from behind with the same relaxed A-pose, arms spread outward and the full silhouette clearly visible. The lower section contains multiple portrait head-angle studies, including upward-looking angles, neutral front-facing views, side and three-quarter views, and downward-looking angles. A larger front-facing portrait appears on the lower right.How to Prompt
The activation sentence should come first.
After that, describe your own subject.
Basic structure:
[Activation Sentence] The subject is [your subject description].Example:
subjectposesheet01. This image is a structured subject reference sheet showing the same subject multiple times in a consistent layout. The top row presents three full-body turnaround views consisting of a front view, a side view, and a back view. In each full-body view, the subject stands in a relaxed A-pose with both arms extended outward. The lower section contains multiple portrait head-angle studies, including upward-looking angles, neutral front-facing views, side and three-quarter views, and downward-looking angles. A larger front-facing portrait appears on the lower right. The subject is a young adult woman with long silver hair, pale blue eyes, fair skin, and a slim athletic build. She is wearing a dark blue futuristic jacket, fitted black trousers, and black boots. The background is a plain dark gray studio backdrop with soft cinematic lighting.Prompting Tips
Describe the subject clearly. This LoRA works best when the subject description is detailed.
Useful things to describe:
body type, age, gender, skin tone, hair, eye color, facial features, clothing, background color, lighting style, realistic style, anime style, illustration styleExample subject description:
The subject is a muscular older man with dark brown skin, short gray hair, broad shoulders, and a calm expression. He is wearing a cream turtleneck sweater, tailored brown trousers, and dark leather shoes. The background is a solid muted green studio backdrop with soft studio lighting.Anime example:
The subject is a stylized anime woman with medium brown skin, bright golden eyes, vivid blue hair, and an athletic build. She is wearing a dark fantasy combat outfit with reinforced shoulder panels, fitted trousers, gloves, and heavy boots. The artwork is rendered as polished anime concept art on a solid dark violet background.Recommended Settings
Model: Krea 2 Raw
Aspect Ratio: 4:3
Resolution: 2 MegapixelsFor best results, keep the activation sentence at the beginning of the prompt and describe your subject after it.
This is an experimental pose-sheet / reference-sheet LoRA. Results may vary depending on prompt length, subject complexity, style, and LoRA strength. Try both activation sentences and adjust your subject description for more control.
Description
Face-Consistent 19-View Character Reference Sheet for Krea 2 Edit
Transform a single character image into a structured 19-view reference sheet while preserving the subject’s recognizable face, appearance, clothing and visual style across the different panels.
Early testing has shown strong facial consistency across the generated views, comparable to the identity retention currently achieved by other character and reference-sheet solutions. The LoRA is designed to keep the same recognizable subject throughout the sheet while generating full-body turnarounds, profile angles and detailed facial studies.
This LoRA was created specifically for the Ostris Krea 2 Edit model and workflow.
What This LoRA Creates
The LoRA transforms a single-person input image into a square character reference sheet containing up to 19 coordinated views:
Three complete full-body turnaround views
Front view
Strict 90-degree side profile
Back view
A consistent relaxed A-pose across the turnaround views
Three horizontal rows containing five head-and-shoulder studies each
Upward-looking head rotation
Neutral-height head rotation
Downward-looking head rotation
One larger front-facing upper-body portrait in the lower-right area
The result is designed as a practical character-development sheet for:
Character design
LoRA dataset preparation
Visual development
Game and animation references
Consistent character planning
Outfit and costume documentation
Character turnaround creation
Face and viewing-angle studies
Face and Character Consistency
Face consistency is one of the main focuses of V02.
The transformation attempts to preserve the following features across every panel:
Recognizable facial identity
Facial anatomy and proportions
Apparent age
Skin tone
Eye shape and general facial structure
Hairstyle and hair color
Body shape and proportions
Clothing and footwear
Accessories and jewelry
Colors, materials and garment construction
Fantasy, alien and species-specific traits
The original rendering style of the input image
Tester feedback indicates that the face remains recognizable and visually coherent across many of the generated angles, including frontal portraits, three-quarter views and side profiles.
Dataset
V02 was trained with 42 paired image-edit examples.
The dataset includes a broad range of subjects and visual styles:
Women and men
Different ages
Different body types
Different skin tones
Photorealistic characters
Fantasy and science-fiction characters
Alien and non-human subjects
Anime and stylized characters
Casual clothing
Formal clothing
Sportswear
Fantasy armor and complex costumes
Bright studio presentations
Dark cinematic presentations
Colored and stylized lighting
Every training caption contains the structural activation tag and detailed instructions for identity preservation, clothing consistency, viewing angles and sheet composition.
The target background and lighting were deliberately varied so that the LoRA would not become tied to one specific presentation style.
Compatibility
Base model: Ostris Krea 2 Edit
Model type: Edit LoRA
Recommended output ratio: 1:1
Suggested starting LoRA strength: 1.0
This LoRA is designed for an image-edit workflow using a single-person reference image.
Activation Tag / Trained Word
AstroSubjectPoseEditV02
Place the activation tag at the beginning of the edit prompt.
Short Activation Sentence
AstroSubjectPoseEditV02. Transform the single-person input image into a structured character reference sheet featuring the exact same subject.
Full Recommended Prompt
AstroSubjectPoseEditV02. Transform the single-person input image into a structured character reference sheet featuring the exact same subject. Treat the input image as the sole visual reference for the subject's identity and appearance, preserving the same facial identity, facial proportions, apparent age, skin tone, hairstyle, hair color, body shape, body proportions, species-specific traits, and overall rendering style. Preserve the exact same clothing, footwear, accessories, colors, materials, and garment construction shown in the input image.
Across the top of the sheet, arrange three complete full-body turnaround views: a front view, a strict 90-degree side profile, and a back view. In every turnaround view, show the entire subject from head to toe in the same relaxed A-pose, with both arms extended outward at a shallow downward angle.
Below the turnaround views, arrange three horizontal rows of five consistent head-and-shoulder studies. The first row shows upward-looking head angles progressing from the left profile through the left three-quarter view, front view, right three-quarter view, and right profile. The second row repeats the same left-to-right rotation at neutral head height. The third row repeats the same rotation while the head looks downward.
Place one larger front-facing upper-body portrait on the lower right. Maintain identical facial anatomy, hairstyle, body proportions, clothing construction, colors, materials, accessories, and visual style across every panel.
Present the complete reference sheet in a square 1:1 composition on a clean warm beige seamless studio background with soft neutral lighting.
The final background and lighting sentence can be freely modified. This makes it possible to create bright studio sheets, neutral technical references, dark cinematic sheets or stylized presentations with colored rim lighting.
Input Recommendations
For the strongest character and face consistency:
Use an input image containing one clearly recognizable subject.
Choose a reference with a clearly visible and sufficiently detailed face.
References showing most of the body and outfit provide more visual information.
Use images with readable clothing shapes, colors and accessories.
Keep the generated reference sheet in a square 1:1 format.
Try multiple seeds to find the strongest individual result.
Keep the complete identity-preservation instructions in the prompt.
Seated, cropped or partially obscured characters can also work. Details that are hidden in the source image will be visually reconstructed by the edit model.
Current Strengths
Strong overall facial recognition across multiple panels
Consistent character appearance
Reliable 19-view sheet structure
Front, side and back full-body reconstruction
Multiple horizontal and vertical head angles
Good preservation of clothing colors and overall design
Support for realistic and stylized characters
Support for fantasy, alien and non-human traits
Flexible background and lighting control
Useful results from a single source image
Alpha Development Note
This release is still officially in Alpha and remains under active development.
Face consistency has already performed well during testing and is generally comparable to the consistency seen in other current character solutions. However, exact identity preservation across every individual panel can never be guaranteed with the current version.
The dataset currently contains 42 paired examples, and the underlying Krea 2 Edit base model can still fluctuate when reconstructing faces, fine accessories, hidden clothing details or extreme viewing angles.
Future versions will expand the dataset and continue improving:
Facial identity preservation
Extreme profile angles
Fine accessory consistency
Hands and feet
Complex fantasy anatomy
Detailed armor and clothing reconstruction
Strict panel placement
I am releasing this Alpha because the current results are already strong and useful enough for real experimentation.
If you test the LoRA, please share your results, settings and feedback. Successful examples as well as difficult inputs are extremely valuable for the next development round.
Have fun experimenting with it!
FAQ
Comments (5)
Hi, just wondering what the use case for character sheets is? for making lora's? for using with reference/edit models?
I have used character sheets for consistentish characters with klein edit but just basic reference sheets with 3 poses, im wondering what the use case here is for?
Thanks
With krea2 you can generate high Resolution Images. After that you crop each Image of a face or Person out and train a LoRA with the multiple Pose and faces for a high consistency. Works really great. The Edit model takes your input image and put it in the upper pose. Why? If you generate a perfect AI woman or man and you want to train a Lora then this Lora can help you with prepare.
@Astroburner Ah cool, That was kind of an idea I had for training lora's,I'm glad to hear it works, I read though that you want different backgrounds etc when training a lora so are you saying you use this as a template to make the images for the training set or you CAN actually just use these sort of images chopped and for building a complete character set to train with... or even a mix of both?
Thanks.
@MrTitsworth You can use the image only single cropped. If you train a Lora with the image it will end to create multiple person with the same look in your Lora. background is ok for the first Lora after that you can generate different backgrounds and train a second Lora for more flexibility
Since you are selling the model now, I assume you trained it on a fully licensed dataset? Have you purchased the rights to use each and every picture used to train this LoRA?