Sign In

Beyond Experimental

84

987

51

Verified:

SafeTensor

Type

Checkpoint Trained

Stats

88

0

Reviews

Published

Aug 9, 2025

Base Model

SDXL 1.0

Hash

AutoV2
B3E7452450

🚀 Project Update – Focus on SDXL Models

We’re making great progress — and I’ve also been updating my 🐴 Pony model. The merging success rate has improved significantly 📈, though more time is needed for further experimentation ⏳.


⚠️ Important Notice

❌ Please do not download version v8. It cannot be deleted at this time, and a re-fix is currently in development 🔧.
📷 This update is primarily focused on image quality improvements.

  • 🎯 v8-Alpha Remastered will be trained at 4K resolution.

  • The current Alpha-8 release runs at 2–3K, and I wasn’t fully satisfied with the results.

💡 4K Training Plan:
To make 4K possible, I’ll need 1–2 months of dedicated training. I’ve already secured the new GPU 💻, but progress will be slow and methodical to ensure the best results.
In the meantime, I’m creating as many LoRA’s as possible 🧪 to test and identify which parameters work best for the model.


🛑 Strategic Shift in Priorities

We are implementing a strategic pause ⏸️ on all other projects to dedicate our full attention to advancing the ✨ SDXL and ⚡ Flux models.

  • 📅 Monthly model updates will be the standard.

  • 💻 Once the GPU upgrade workflow is fully in place, we’ll move to bi-weekly updates starting with the Alpha 2.5 release.

  • 🏋️ Following that, we’ll dedicate two full months of intensive training to ensure unparalleled quality.

(🎬 Yes — those amazing videos you’ve seen were created using Veo 3!)


🔙 BeyondNegative is Back!

We’re excited to confirm that BeyondNegative’s trigger word functionality will be fully restored and operational in Alpha 0.6 ✅

Key details:

  • 🚫 BeyondNegative is not an embedding or LoRA — it’s an integrated negative trigger word.

  • 📌 Do not place it in the positive prompt.

  • 🧩 Most negative embeddings form the backbone of the unreleased Beyond Experimental v3.2 and use special blocks.

  • 🛠️ New negative embeddings are already in active development.


🔗 Merging Compatibility

  • 💥 Will break if merged with any Pony or Illustrious models.

  • ⚠️ Partial compatibility with Rog Pony or Anime versions — but success rate is low.

  • 💾 Using the FP32 model, BeyondNegative should still function since it was trained on both.

  • ❗ However, Beyond Experimental is not fully compatible with either Pony or Illustrious.


📥 Download:
👉 BadPonyUHD_BETA – Hugging Face Link

🙏 Thank you for your patience, understanding, and continued support! Your feedback keeps us pushing forward. ❤️

📢 Upcoming Updates

In future releases, Source_anime and Source_realistic will be added as official trigger words.


⚠️ Important for LoRA Creators

When creating LoRAs for this model, you must always place Source_anime or Source_realistic at the beginning of the prompt during training.

Failing to do so will result in overwriting the trigger words, which will break intended functionality and reduce compatibility.
While creators are free to ignore this guideline and work however they wish, doing so will make it significantly more difficult for others to use the model effectively.


🆔 Final Model Name

When the model reaches full release, it will be officially titled: BeyondSDXL.


🔍 Merging & Verification

If you merge this model with another, please provide:

  1. An XYZ plot showing the merge results.

  2. The same prompt and seed for each comparison.

This ensures users can verify whether:

  • The merge was successful ✅

  • The model remains functional 🔧

  • No parts were stolen from another creator ❌