What is Mixpeek?

[Back to Blog](/blog)

[](https://twitter.com/intent/tweet?text=Multimodal%20Monday%20AI%20%2337%3A%20Less-Is-More%2C%20Analogical%20Vision&url=https%3A%2F%2Fmixpeek.com%2Fblog%2Fmultimodal-monday-37-less-is-more-analogical-vision)[](https://www.linkedin.com/shareArticle?mini=true&url=https%3A%2F%2Fmixpeek.com%2Fblog%2Fmultimodal-monday-37-less-is-more-analogical-vision&title=Multimodal%20Monday%20AI%20%2337%3A%20Less-Is-More%2C%20Analogical%20Vision)[](https://www.reddit.com/submit?url=https%3A%2F%2Fmixpeek.com%2Fblog%2Fmultimodal-monday-37-less-is-more-analogical-vision&title=Multimodal%20Monday%20AI%20%2337%3A%20Less-Is-More%2C%20Analogical%20Vision)

This was originally in our newsletter.

Get Updates

Dec 15, 2025•6 min read

# Multimodal Monday AI #37: Less-Is-More, Analogical Vision

Week of Dec 9-14, 2025: Apple proves one attention layer beats dozens in diffusion models, MokA shows low-rank adaptation outperforms full fine-tuning, Adobe's relsim captures analogical relationships between images, and X-VLA controls different robot types with one transformer.

![Multimodal Monday AI #37: Less-Is-More, Analogical Vision](https://blog.mixpeek.com/content/images/2025/12/mm_37.png)

Multimodal Monday

TABLE OF CONTENTS

### 📢 Quick Hits (TL;DR)

**AI understands analogies between images.** Adobe’s new relsim metric helps models grasp deep visual relationships. A peach’s cross-section mirrors Earth’s layers. Your model can finally see these connections.

**Apple proves one attention layer can beat dozens.** Their new paper shows a single attention layer converts pretrained vision features into SOTA image generators. MokA shows you can adapt multimodal models with fewer parameters and better results.

**AI gets a body (and a phone).** X-VLA is a soft-prompted transformer that controls totally different robots like they’re one species. And AutoGLM, open-source phone agent framework, enables autonomous task completion on Android devices through natural language commands.

### 🧠 Research Highlights

#### **Relational Visual Similarity**

Adobe Research and UW-Madison developed relsim, a metric that captures analogical relationships between images rather than surface-level features. The model understands that a peach’s layers relate to Earth’s structure the same way a key relates to a lock.  
**Links**: [Paper](https://www.arxiv.org/pdf/2512.07833?ref=blog.mixpeek.com)

 Your browser does not support the video tag.

#### **One Attention Layer is Enough**

Apple demonstrates that a single attention layer transforms pretrained vision features into state-of-the-art image generators. This approach simplifies diffusion models while maintaining top-tier quality.  
**Links**: [Paper](https://arxiv.org/abs/2512.07829?ref=blog.mixpeek.com)

![](https://substackcdn.com/image/fetch/$s_!bYyF!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fc11f3b95-e29b-4a6e-b17b-22702e987059_2294x1012.jpeg)

Comparison between standard VAE (Rombach et al., 2022b), VA-VAE (Yao et al., 2025), RAE (Zheng et al., 2025) and their proposed FAE. The number shows the channel dimension of the generative modeling space.

#### **Unison: A Fully Automatic, Task-Universal, and Low-Cost Multimodal Framework**

Unison automates multimodal tasks across text, images, and video without task-specific retraining. The framework uses efficient fusion techniques that work across different modalities with minimal computational overhead.  
**Links**: [Paper](https://arxiv.org/html/2512.07747v1?ref=blog.mixpeek.com)

![](https://substackcdn.com/image/fetch/$s_!OxMr!,w_2400,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F928ef70e-ca9f-473e-9084-d2173526d767_997x392.png)

Overview of the Unison. The left shows the stage-one understanding model, the right shows the stage-two generation model. These two components are connected via a projector.

**MokA: Multimodal Low-Rank Adaptation for MLLMs**

MokA reveals that current multimodal fine-tuning wastes parameters and proposes a low-rank method that improves visual-language integration. The approach beats standard fine-tuning on visual grounding benchmarks while using fewer parameters.  
**Links**: [Paper](https://arxiv.org/html/2506.05191v2?ref=blog.mixpeek.com)

![](https://substackcdn.com/image/fetch/$s_!bWp4!,w_2400,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F1627cc0c-e6e8-48c9-ac29-41f7ad466215_1520x1320.jpeg)

Evaluation results of LoRA, its variants, and MokA on visual-text benchmarks. #A and #B are the number of low-rank matrices. Here N=2 refers to the number of modalities.

**Computational Emotion Analysis with Multimodal LLMs**

Researchers evaluate how well multimodal LLMs analyze emotions from audio-visual data. The paper identifies gaps in current methods and shows where generative AI can improve affective computing.  
**Links**: [Paper](https://arxiv.org/abs/2512.10882?ref=blog.mixpeek.com)

![](https://substackcdn.com/image/fetch/$s_!HTq7!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd0946192-05f1-4a35-b8b8-68db45e74787_1954x490.jpeg)

Relationship between video-based 3-shot emotion intensity scores of mLLMs and human coders’ average ratings in the RAVDESS dataset

**WonderZoom:** Generates multi-scale 3D worlds from text descriptions. Links: [Paper](https://huggingface.co/papers/2512.09164?ref=blog.mixpeek.com)

 Your browser does not support the video tag.

**X-VLA:** Soft-prompted transformer that controls different robot types using a unified interface. Links: [Docs](https://huggingface.co/docs/lerobot/en/xvla?ref=blog.mixpeek.com)

![](https://substackcdn.com/image/fetch/$s_!Eg8p!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0a6dc7b9-eddf-4d82-9daf-6a4a4b807893_900x437.png)

**MoCapAnything:** Captures 3D motion for arbitrary skeletons from single-camera videos. Links: [Paper](https://huggingface.co/papers/2512.10881?ref=blog.mixpeek.com)

 Your browser does not support the video tag.

**RouteRAG:** Uses reinforcement learning to efficiently retrieve from both text documents and knowledge graphs. Links: [Paper](https://arxiv.org/abs/2512.09487?ref=blog.mixpeek.com) | [GitHub](https://github.com/YucanGuo/RouteRAG?ref=blog.mixpeek.com)

### 🛠️ Tools, Models and Techniques

#### **AutoGLM**

z.ai released AutoGLM, an open-source framework that completes tasks on Android phones through natural language commands. AutoGLM-Phone-9B is available on Hugging Face and ModelScope.

 Your browser does not support the video tag.

Why it matters: You can now build agents that interact with real-world smartphone apps.  
**Links**: [Website](https://xiao9905.github.io/AutoGLM/?ref=blog.mixpeek.com)

#### **Apriel-1.6-15B-Thinker**

ServiceNow’s 15B multimodal model scores 57 on the Artificial Analysis Intelligence Index, matching the performance of 200B-scale models. The model handles reasoning tasks at a fraction of the size.

## Run this on your own content

Point Mixpeek at the storage you already have and search your video, images, audio, and documents the way this post describes. Build starts at $25/mo for up to 1M vectors.

[Search your own files](https://studio.mixpeek.com/namespaces/create?mode=managed&utm_source=homepage&utm_medium=blog_onramp_mid&utm_campaign=mvs_onramp)[Talk to us](/demo)

![](https://substackcdn.com/image/fetch/$s_!2guv!,w_2400,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ffd2b3a82-4ca7-4d2e-a213-cbf0a2c2605e_800x298.jpeg)

Why it matters: You get frontier-level multimodal reasoning in a model small enough to self-host.  
**Links**: [Model](https://huggingface.co/ServiceNow-AI/Apriel-1.6-15b-Thinker?ref=blog.mixpeek.com) | [Blog](https://huggingface.co/blog/ServiceNow-AI/apriel-1p6-15b-thinker?ref=blog.mixpeek.com) | [Demo](https://huggingface.co/spaces/ServiceNow-AI/Apriel-Chat?ref=blog.mixpeek.com)

#### **GLM-4.6V**

Z.ai released GLM-4.6V with tool-calling and 128K context window for vision-language tasks. The model handles multilingual development and API integration.

![](https://substackcdn.com/image/fetch/$s_!voxd!,w_2400,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7dce24f7-0bd3-4e84-aeab-00d85f7bb1dc_10101x7371.jpeg)

Why it matters: You get long-context multimodal processing with native tool support.  
**Links**: [Blog](https://z.ai/blog/glm-4.6v?ref=blog.mixpeek.com) | [GitHub](https://github.com/zai-org/GLM-V?ref=blog.mixpeek.com) | [Demo](https://chat.z.ai/?ref=blog.mixpeek.com)

#### **GPT-5.2**

OpenAI released GPT-5.2, their latest frontier model. The model advances capabilities across reasoning, generation, and multimodal understanding.

Why it matters: The new baseline for frontier model performance just shifted.  
**Links**: [Twitter](https://x.com/OpenAI/status/1999182098859700363?s=20&ref=blog.mixpeek.com) | [Blog](https://openai.com/index/introducing-gpt-5-2/?ref=blog.mixpeek.com)

#### **DMVAE**

Tencent and PKU released DMVAE, a VAE that matches latent distributions to any reference. The model achieves state-of-the-art image synthesis with fewer training epochs.

![](https://substackcdn.com/image/fetch/$s_!V5V7!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fd9163d7f-823f-4f59-a04e-7765adc8bd79_692x900.jpeg)

Why it matters: You can control the distribution of your generated images more precisely.  
**Links**: [Paper](https://huggingface.co/papers/2512.07778?ref=blog.mixpeek.com) | [Model](https://huggingface.co/sen-ye/dmvae/tree/main?ref=blog.mixpeek.com)

**Qwen-Image-i2L:** First open-source tool that converts a single image into a custom LoRA. Links: [ModelScope](https://modelscope.cn/models/DiffSynth-Studio/Qwen-Image-i2L/summary?ref=blog.mixpeek.com) | [Code](https://github.com/modelscope/DiffSynth-Studio/blob/main/examples/qwen_image/model_inference_low_vram/Qwen-Image-i2L.py?ref=blog.mixpeek.com)

**Dolphin-v2:** 3B parameter model that parses any document type. Links: [Hugging Face](https://huggingface.co/ByteDance/Dolphin-v2?ref=blog.mixpeek.com)

**RealGen:** Improves text-to-image photorealism using detector-guided rewards. Links: [Website](https://yejy53.github.io/RealGen/?ref=blog.mixpeek.com) | [Paper](https://huggingface.co/papers/2512.00473?ref=blog.mixpeek.com) | [GitHub](https://github.com/yejy53/RealGen?tab=readme-ov-file&ref=blog.mixpeek.com) | [Models](https://huggingface.co/lokiz666/Realgen-detection-models?ref=blog.mixpeek.com)

![](https://substackcdn.com/image/fetch/$s_!zfp9!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F54d07455-8eeb-4d84-9ffd-82b89daf8a74_1200x501.jpeg)

**Qwen 360 Diffusion:** State-of-the-art 360° text-to-image generation. Links: [Hugging Face](https://huggingface.co/ProGamerGov/qwen-360-diffusion?ref=blog.mixpeek.com) | [Viewer](https://progamergov.github.io/html-360-viewer/?ref=blog.mixpeek.com)

![](https://substackcdn.com/image/fetch/$s_!PPJR!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa287d942-5389-4261-a9de-14d9ac357a54_12000x10000.jpeg)

**Any4D:** Unified transformer for feed-forward, dense, metric-scale 4D reconstruction. Links: [Website](https://any-4d.github.io/?ref=blog.mixpeek.com) | [Paper](https://any-4d.github.io/assets/Any4D.pdf?ref=blog.mixpeek.com) | [Demo](https://huggingface.co/spaces/theairlabcmu/Any4D?ref=blog.mixpeek.com)

 Your browser does not support the video tag.

**Shots:** Generates 9 cinematic camera angles from a single image with perfect consistency. Links: [Twitter](https://x.com/higgsfield_ai/status/1998895357707825503?s=20&ref=blog.mixpeek.com)

 Your browser does not support the video tag.

### 📈 Trends & Predictions

#### **The Great Simplification**

We’re watching complexity collapse in real time. Apple needs one attention layer where others use dozens. Unison handles any modality without retraining. MokA beats full fine-tuning with a fraction of the parameters.

This isn’t about making things smaller. It’s about removing the cruft that never belonged there.

##### **What This Means for You**

Your multimodal systems can run faster and cheaper. A single-layer model fits on hardware that was out of reach last year. You can index a billion images on a single GPU.

Your systems can become more adaptable. Unison adds new modalities without retraining. You can build applications that were too complex before.

Your models become more transparent. Low-rank adaptation means fewer parameters to debug. You can explain why your system retrieved a specific image.

This is multimodal AI for teams with real constraints. Not just for companies with unlimited compute budgets.

### 🧩 Community + Shoutouts

**Nano Banana Pro Solution** Shoutout to rob from ComfyUI for sharing a solution that addresses the biggest downside to Nano Banana Pro (the cost and slow generation speed). 1 prompt = 9 distinct images at 1K resolution (~3 cents per image). Links: [Twitter](https://x.com/hellorob/status/1999537115168636963?s=42&ref=blog.mixpeek.com)

 Your browser does not support the video tag.

**Multimodal Benchmarks GitHub Repo** Shoutout to Mixpeek for releasing a Multimodal Benchmarks GitHub repo. A great resource for anyone who is working on multimodal AI. Links: [GitHub](https://github.com/mixpeek/multimodal-benchmarks?ref=blog.mixpeek.com)

**67 New Z Image LoRAs** Shoutout to Reddit user malcolmrey for releasing 67 new Z Image LoRAs. A great resource for anyone who is using Z Image. Links: [Reddit](https://www.reddit.com/r/malcolmrey/comments/1pgv65v/67_new_z_image_loras/?ref=blog.mixpeek.com)

* * *

**That’s a wrap for Multimodal Monday #37!** Apple proves one attention layer generates SOTA images. Adobe’s relsim captures analogies between images. MokA beats full fine-tuning with low-rank adaptation. RouteRAG navigates knowledge graphs with reinforcement learning. X-VLA controls different robots with one transformer. AutoGLM completes tasks on Android phones. Apriel-1.6-15B matches 200B-scale models at 15B parameters.

_Ready to build multimodal solutions that actually work?_ [_Let's talk_](https://mixpeek.com/contact?ref=blog.mixpeek.com)

[PB

Philip Bankier

Philip’s expertise in RAG, data engineering, and AI agents makes him a natural fit to write about scaling multimodal data warehousing systems.



](/blog/authors/philip)

## Run this on your own content

Point Mixpeek at the storage you already have and search your video, images, audio, and documents the way this post describes. Build starts at $25/mo for up to 1M vectors.

[Search your own files](https://studio.mixpeek.com/namespaces/create?mode=managed&utm_source=homepage&utm_medium=blog_onramp_end&utm_campaign=mvs_onramp)[Talk to us](/demo)

## Continue Learning

[

![Multimodal Monday #45: Birds, Whales, and the End of Latency](https://blog.mixpeek.com/content/images/2026/02/mm_45.png)

### Multimodal Monday #45: Birds, Whales, and the End of Latency

Your Weekly Multimodal AI Roundup (Feb 9 - Feb 16)

Read More





](/blog/multimodal-monday-45)[

![Multimodal Monday #44: Agents Ship Code, Robots Skateboard](https://blog.mixpeek.com/content/images/2026/02/mm44---with-milo.png)

### Multimodal Monday #44: Agents Ship Code, Robots Skateboard

Feb 2 - 9: GPT-5.3-Codex and Claude Opus 4.6 handle full software lifecycles from debug to deploy, MiniCPM-o 4.5 beats GPT-4o on vision tasks at 9B parameters running on-device, HUSKY skateboards using physics-based control, and TinyLoRA fine-tunes models with a single parameter.

Read More





](/blog/multimodal-monday-44)

### Author

PB

Philip Bankier

![Mixpeek Icon](/brand/logo-dark-icon.webp)

#### Multimodal Search, On Demand

Index, retrieve, and reason over videos, images, audio, and docs, all from a single API.

[Start Building](/start)

### Share

[](https://twitter.com/intent/tweet?text=Multimodal%20Monday%20AI%20%2337%3A%20Less-Is-More%2C%20Analogical%20Vision&url=https%3A%2F%2Fmixpeek.com%2Fblog%2Fmultimodal-monday-37-less-is-more-analogical-vision)[](https://www.linkedin.com/shareArticle?mini=true&url=https%3A%2F%2Fmixpeek.com%2Fblog%2Fmultimodal-monday-37-less-is-more-analogical-vision&title=Multimodal%20Monday%20AI%20%2337%3A%20Less-Is-More%2C%20Analogical%20Vision)[](https://www.reddit.com/submit?url=https%3A%2F%2Fmixpeek.com%2Fblog%2Fmultimodal-monday-37-less-is-more-analogical-vision&title=Multimodal%20Monday%20AI%20%2337%3A%20Less-Is-More%2C%20Analogical%20Vision)