HorizonSphere XL — Core

HorizonSphere XL — Core is an experimental local-first cinematic checkpoint built around controlled composition, immersive horizon design, HDRI / 360° environment direction, hyperreal detail, African fashion influence, character staging, and modular SDXL-style LoRA support.

This release is part of the HorizonSphere XL series: a model family focused on cinematic framing, spherical worldbuilding, panoramic scene design, editorial realism, and production-oriented ComfyUI use.

TECHNICAL NOTE:

This model should not be described as a normal pure SDXL checkpoint if the supplied merge recipe is accurate.

The recipe includes SDXL components, SDXL LoRAs, SDXL text encoders, and an SDXL VAE, but the dominant checkpoint contribution is a FLUX.2 Klein 9B derivative. Treat this as an experimental hybrid / Klein-oriented XL merge unless local validation proves a different architecture and loading path.

MODEL IDENTITY

Public model name: HorizonSphere XL — Core

Series name: HorizonSphere XL

Release role: Core foundation release

Primary aesthetic target:

Cinematic realism, panoramic worldbuilding, fashion-led portraiture, HDRI-style lighting, character identity, and high-detail editorial rendering.

“Horizon” refers to cinematic framing, scene stability, atmospheric depth, and controlled environmental perspective.

“Sphere” refers to immersive 360° skybox direction, equirectangular panorama testing, and spherical HDRI-style worldbuilding.

“XL” identifies the production intent around large-format SDXL-style image generation, while preserving the technical warning that this recipe uses Klein / Flux-family assets.

CORE CONCEPT

HorizonSphere XL — Core is designed for creators who want a single foundation checkpoint for cinematic scenes, stylized realism, controlled camera direction, 360° environment studies, editorial portraiture, and modular LoRA expansion.

The release is built around a small merge value from a personal SDXL merge, small reinforcement from a refined Flux2Klein 4B base, and a dominant Flux2Klein 9B BigLove Klein 2 FP8 pruned component.

The LoRA layer then introduces fashion, HDRI, maternity detail, automatic detail enhancement, character identity, and hyperreal cinematic rendering.

CHECKPOINT MERGE RECIPE

The following recipe is documented for transparency. It is not a recommendation to mix incompatible architectures unless your merge tool and runtime path explicitly support the operation.

1.

Asset: sdxl_20260419_METATRON.safetensors

Role: Personal SDXL merge model

Merge value: 0.01

Source / credit: Private / personal merge

2.

Asset: flux2klein-4b-base_20260410_refined_alpha1_7gb.safetensors

Role: Flux2Klein 4B refined base contribution, used two times

Merge value: 0.01 each

Source / credit: https://civarchive.com/models/2530956/f2k4brefinedbase

3.

Asset: flux2klein-9b_20260425_big_love_klein2_fp8_pruned_1_9gb.safetensors

Role: Dominant Flux2Klein / BigLove Klein 2 FP8 pruned contribution

Merge value: 0.98

Source / credit: https://huggingface.co/Granddyser/biglove-klein2/blob/main/bigLove_klein2_fp8_pruned.safetensors

The BigLove Klein 2 source page describes the FP8 pruned file as a roughly 9 GB variant intended as a balance between quality and VRAM. The page also identifies the model family as FLUX.2 Klein 9B rather than standard SDXL. For this reason, users should not assume that every SDXL workflow, sampler path, text encoder, or loader will work without validation.

LORA STACK

The LoRA layer is intentionally light. Each listed LoRA is documented at 0.09, which keeps the stack subtle and reduces the risk of overcooking the base model.

1.

LoRA file: lora_sdxl_20240417_concept_african_fashion_1.0.safetensors

Value: 0.09

Purpose: African fashion concept, textile identity, wardrobe direction, cultural styling

Credit / source: https://civarchive.com/models/509626/concept-african-fashion

2.

LoRA file: sdxl_lora_20240613_tool_hdri_360_degree_skybox_creator.safetensors

Value: 0.09

Purpose: HDRI / 360° skybox creation, spherical environments, panoramic view direction

Credit / source: https://civarchive.com/models/492184/hdri-360-degree-skybox-creator-for-sdxl

3.

LoRA file: lora_sdxl_20240324_pregnant_detailer_v1.safetensors

Value: 0.09

Purpose: Respectful adult maternity detail support

Credit / source: https://civarchive.com/models/364811/pregnant-detailer-sdxl

4.

LoRA file: lora_sdxl_20260420_concept_adetailer_v6.safetensors

Value: 0.09

Purpose: Detail enhancement concept, face / hand / local refinement support

Credit / source: https://civarchive.com/models/2559146/adetailerxl

5.

LoRA file: sdxl_lora_20250628_character_Marie_LAVEAU_epoch_1.safetensors

Value: 0.09

Purpose: Character identity and historical / symbolic visual direction

Credit / source: https://civarchive.com/models/1724242/character-marie-laveau

6.

LoRA file: lora_sdxl_20250814_style_hyper_realism_v1.safetensors

Value: 0.09

Purpose: Hyperrealism, cinematic detail, material realism, true-to-life rendering

Credit / source: https://civarchive.com/models/1865288/hyperrealism-cinematic-detail-and-true-to-life-rendering-sdxl

Recommended first-use policy:

Test the base model alone, then test one LoRA at a time, then test the full stack. If the image becomes overworked, reduce detail LoRA influence before changing the sampler or negative prompt.

TEXT ENCODERS

The declared text encoder stack uses an SDXL-style CLIP-G / CLIP-L pairing.

1.

Text encoder file: sdxl_text-encoder_bigonly_bigaspv_2_clip_g.safetensors

Role: CLIP-G / text_encoder_2 style SDXL encoder

Credit / source: https://huggingface.co/John6666/big-asp-v2-sdxl/tree/main/text_encoder_2

2.

Text encoder file: sdxl_text-encoder_clip_l.safetensors

Role: CLIP-L / text_encoder style SDXL encoder

Credit / source: https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/blob/main/text_encoder/model.safetensors

The John6666 Big ASP v2 SDXL repository publicly exposes SDXL-style folders including text_encoder, text_encoder_2, tokenizer, tokenizer_2, unet, and vae. The Stability AI SDXL base page also identifies SDXL as using two fixed pretrained text encoders. This supports the SDXL-style text encoder labeling, but it does not automatically solve compatibility if the dominant model body is Flux2Klein / Klein 9B.

VAE

Declared VAE: sdxl_night-photo-hdr_base.safetensors

Source: https://civarchive.com/models/1480240/hdr-effect-vae

Intended role:

Night-photo HDR effect, richer tonal response, stronger luminous atmosphere, cinematic highlight behavior, and HDR-oriented image decoding.

Compatibility note:

If the runtime path is genuinely Flux2Klein-oriented, verify that this VAE is valid for the selected loader and model architecture. If the workflow loads as SDXL, place the VAE in the normal ComfyUI VAE folder and manually select it in the decode path.

Recommended VAE path:

ComfyUI/models/vae/sdxl_night-photo-hdr_base.safetensors

COMFYUI PLACEMENT

Because this recipe includes both SDXL and Flux2Klein assets, do not assume one folder route is correct for all files. Use the loader required by the actual architecture that successfully imports.

For the personal SDXL checkpoint component:

ComfyUI/models/checkpoints/sdxl_20260419_METATRON.safetensors

For Flux2Klein / BigLove Klein diffusion model files:

ComfyUI/models/diffusion_models/flux2klein-4b- base_20260410_refined_alpha1_7gb.safetensors

ComfyUI/models/diffusion_models/flux2klein-9b_20260425_big_love_klein2_fp8_pruned_1_9gb.safetensors

For LoRAs:

ComfyUI/models/loras/lora_sdxl_20240417_concept_african_fashion_1.0.safetensors

ComfyUI/models/loras/sdxl_lora_20240613_tool_hdri_360_degree_skybox_creator.safetensors

ComfyUI/models/loras/lora_sdxl_20240324_pregnant_detailer_v1.safetensors

ComfyUI/models/loras/lora_sdxl_20260420_concept_adetailer_v6.safetensors

ComfyUI/models/loras/sdxl_lora_20250628_character_Marie_LAVEAU_epoch_1.safetensors

ComfyUI/models/loras/lora_sdxl_20250814_style_hyper_realism_v1.safetensors

For text encoders:

ComfyUI/models/text_encoders/sdxl_text-encoder_bigonly_bigaspv_2_clip_g.safetensors

ComfyUI/models/text_encoders/sdxl_text-encoder_clip_l.safetensors

For VAE:

ComfyUI/models/vae/sdxl_night-photo-hdr_base.safetensors

Validation rule:

The model is not production-ready until it imports, resolves all assets, reaches VAE decode or the equivalent terminal output, and produces a saved image without architecture, text-encoder, or tensor shape mismatch errors.

RECOMMENDED GENERATION SETTINGS

These settings are conservative starting points. They must be validated locally because the checkpoint recipe is hybrid and may not behave like a conventional SDXL checkpoint.

Core Cinematic Still:

Resolution: 1024x1024, 1152x896, 896x1152, 1216x832, or 832x1216

Sampler: Use the sampler required by the working loader

Steps: 24–36 for SDXL-style routes

CFG: 4.5–7.0 for SDXL-style routes

Batch size: 1

Seed: Fixed for comparison, variable for exploration

VAE: sdxl_night-photo-hdr_base.safetensors if compatible

Flux2Klein / BigLove Klein Oriented Smoke Test:

Resolution: 768x768 or 1024x1024

Steps: Start low and follow the source workflow defaults

Guidance / CFG: Follow the Flux2Klein loader expectation

Batch size: 1

Upscaler: Disabled during first validation

Parallel branches: Disabled during first validation

HDRI / 360° Panorama Test:

Aspect ratio: 2:1

Resolution examples: 2048x1024, 1536x768, 1344x672, 1024x512

Prompt target: Equirectangular 360 degree panorama

Validation:

Test the image in a 360° panorama viewer:

https://renderstuff.com/tools/360-panorama-web-viewer/?utm_source=chatgpt.com

SHOWCASE PROMPT — HORIZONSPHERE XL CORE

Use this prompt for a first Civitai showcase image:

equirectangular 360 degree panorama, cinematic HDRI skybox environment, seamless spherical world, 2:1 panoramic composition, centered horizon line, immersive Afro-futurist desert palace at sunset, elegant African fashion-inspired architectural patterns, golden hour lighting, soft volumetric clouds, distant mountains, detailed stone and bronze structures, atmospheric dust, realistic sky gradient, balanced exposure, natural global illumination, high dynamic range lighting, ultra-wide panoramic view, no visible seam, continuous left-right horizon, clean sky-to-ground transition, professional virtual production background, photorealistic, hyperreal cinematic detail, true-to-life rendering, sharp environmental detail, no people, no text, no logo

NEGATIVE PROMPT

visible seam, broken horizon, duplicated sun, distorted sky, warped architecture, bad perspective, vertical split, mismatched left and right edges, black borders, fisheye artifacts, blurry, low quality, text, watermark, logo, people, faces, bodies, vehicles, cropped panorama, non-equirectangular image, bad tiling, broken symmetry, overexposed sky, muddy shadows

SHOWCASE CAPTION

HorizonSphere XL — Core showcase image. Experimental local-first cinematic XL / Klein hybrid merge focused on panoramic worldbuilding, HDRI-style lighting, African fashion influence, controlled horizon design, and hyperreal editorial atmosphere. The output is designed as an equirectangular 2:1 panorama and can be tested in a 360° viewer for seam continuity, horizon stability, immersive rotation, and skybox usability.

HDRI / 360° PANORAMA TEST VIEWER

Test generated HDRI / 360° panoramic views here:

https://renderstuff.com/tools/360-panorama-web-viewer/?utm_source=chatgpt.com

The RenderStuff viewer accepts equirectangular panoramic images, displays them interactively, supports rotate / zoom / fullscreen viewing, and can generate shareable URLs or embed code. Local images stay in the browser unless you intentionally use a hosted image URL for sharing.

Recommended checks:

- 2:1 equirectangular aspect ratio

- left/right seam continuity

- centered and stable horizon line

- no duplicated sun or broken lighting direction

- no vertical split

- no warped architecture

- clean sky-ground transition

- usable immersive rotation

- no visible generation break at the poles

PROMPT TEMPLATE

cinematic editorial image, controlled horizon design, immersive environment, realistic global illumination, high dynamic range lighting, detailed material rendering, hyperreal cinematic detail, true-to-life rendering, professional composition, atmospheric depth, [subject or environment], [camera direction], [lighting], [wardrobe or architecture], [mood], [LoRA-specific trigger words if required]

GENERAL NEGATIVE PROMPT

worst quality, low quality, blurry, bad anatomy, deformed hands, extra fingers, missing fingers, distorted face, plastic skin, overprocessed, oversharpened, bad proportions, duplicate body, text, watermark, logo, jpeg artifacts, broken lighting, flat depth, malformed architecture

RESPONSIBLE USE

This model stack is intended for responsible adult and cinematic image generation. When using character, pregnancy, historical, religious, culturally specific, or identity-related LoRAs, prompt with respect and avoid exploitative framing.

For maternity-related imagery, keep all subjects adult, non-explicit, and respectfully represented.

For historical or culturally inspired character work, avoid caricature and use contextual visual language: wardrobe, era, place, lighting, symbolic objects, and atmosphere.

KNOWN TECHNICAL LIMITS

HorizonSphere XL — Core is an experimental architecture-sensitive release. Users should validate the loader path before assuming compatibility with standard SDXL tools.

Known limits:

- The recipe includes both SDXL and Flux2Klein / Klein-family assets.

- The dominant contribution is a Flux2Klein 9B / BigLove Klein 2 FP8 pruned model.

- SDXL CLIP-L / CLIP-G text encoders may not be valid for a Flux2Klein-native runtime path.

- SDXL LoRAs may not apply correctly to non-SDXL architectures unless the merge process has already internalized their effect or a compatible adapter route exists.

- The VAE must be verified in the actual workflow path before publishing fixed settings.

Do not mark the model as production-stable until local import, asset resolution, sampler execution, decode, and output save have all passed.

PARTNERS, CREDITS, AND OPEN-SOURCE RESOURCES

This release exists thanks to the wider open-source and creator ecosystem around SDXL, Flux2Klein, ComfyUI, Civitai, Hugging Face, Stability AI, LoRA authors, VAE publishers, and tool builders. Please visit, support, credit, and respect the licenses of the original creators.

CHECKPOINT / BASE RESOURCES

Personal SDXL merge: sdxl_20260419_METATRON.safetensors

F2K 4B Refined Base: https://civarchive.com/models/2530956/f2k4brefinedbase

BigLove Klein 2 FP8 Pruned: https://huggingface.co/Granddyser/biglove-klein2/blob/main/bigLove_klein2_fp8_pruned.safetensors

LORA RESOURCES

African Fashion: https://civarchive.com/models/509626/concept-african-fashion

HDRI 360 Degree Skybox Creator for SDXL: https://civarchive.com/models/492184/hdri-360-degree-skybox-creator-for-sdxl

Pregnant Detailer SDXL: https://civarchive.com/models/364811/pregnant-detailer-sdxl

ADetailerXL: https://civarchive.com/models/2559146/adetailerxl

Character Marie Laveau: https://civarchive.com/models/1724242/character-marie-laveau

Hyperrealism, Cinematic Detail and True-to-Life Rendering SDXL: https://civarchive.com/models/1865288/hyperrealism-cinematic-detail-and-true-to-life-rendering-sdxl

TEXT ENCODER RESOURCES

Big ASP v2 SDXL text_encoder_2: https://huggingface.co/John6666/big-asp-v2-sdxl/tree/main/text_encoder_2

Stability AI SDXL CLIP-L text encoder: https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/blob/main/text_encoder/model.safetensors

VAE RESOURCE

HDR Effect VAE: https://civarchive.com/models/1480240/hdr-effect-vae

HDRI / 360° TEST VIEWER

RenderStuff Online 360° Panorama Viewer: https://renderstuff.com/tools/360-panorama-web-viewer/?utm_source=chatgpt.com

ATTRIBUTION NOTE

All linked resources remain the property of their respective authors and platforms. This release does not claim ownership over partner checkpoints, LoRAs, text encoders, VAEs, tools, or external model resources.

Users should check each linked page for license terms, usage permissions, version notes, creator instructions, commercial restrictions, redistribution rules, and merge permissions before commercial use, redistribution, merging, or derivative publishing.

RECOMMENDED TAGS

HorizonSphere XL, SDXL, Flux2Klein, Klein, cinematic, realism, hyperrealism, HDRI, 360 panorama, skybox, African fashion, editorial, ComfyUI, local-first, experimental merge, worldbuilding

FINAL POSITIONING

HorizonSphere XL — Core is the foundation release of the HorizonSphere XL series: an experimental cinematic merge focused on horizon control, spherical environment design, HDRI-style worldbuilding, hyperreal detail, and modular creator attribution.

It is built for serious testing, transparent publishing, and local-first creative production.

Files with SHA256: cd6e6df378f473379faaeaaee1d35a881588e46abd14b8e027ea4df0711ac399