Creative Visual Generation Editing Understanding

🎉We are hosting the 7th edition of our CVEU Workshop at #SIGGRAPH2025 ! @siggraph Join us in Vancouver for a half-day tour exploring endless possibilities in AI for creative visual content. Stay tuned for further details of the event at cveu.github.io 🤩🤩

Deadline Update: Following news from the CVPR workshop chairs, we have revised the In-Proceedings Track deadlines: 📅 Abstract Registration: Mar 16, 2026 📄 Submission Deadline: Mar 17, 2026 Visit our website for full details and submission info: 🔗 cveu.github.io

📢 𝑪𝒂𝒍𝒍 𝒇𝒐𝒓 𝑷𝒂𝒑𝒆𝒓𝒔: @CVPR 2026 Workshop on AI for Creative Visual Content Generation, Editing, and Understanding (CVEU)! We are thrilled to invite submissions to the 9th installment of the CVEU workshop at CVPR 2026 in Denver!

The day has arrived! #CVPR2026 reviews are scheduled to be released sometime today. NO, we don’t know the exact release time, we’ll find out together with the community. Only our Program Chairs do. Good luck to everyone! 🤞

@mariyaivasileva @gabriberton Was it @JitendraMalikCV who once lamented the loss of careful scholarship: papers citing mostly post-2015/2017 work, superficial related-work sections, and a failure to build cumulatively on decades of prior insights?

My general principle is something along the lines of, “if it’s in textbooks or a widely-adopted concept, no need to cite original paper(s)”. But here I’m referencing the all-too-common, “let’s throw the off-the-shelf LLM hammer at this problem” without comparing to some domain-specific model because it wasn’t published in the last couple of years (yet, there wasn’t an alternative SOTA reference either).

Feeling a strange mix of nostalgia and confusion after wrapping up this CVPR paper review cycle. This is probably a non-exhaustive list, but here are some raw observations of themes I saw repeat across submissions. - Extreme recency bias: references rarely leave the realm of the 2020s. - Increasingly “single-context” evaluation, rather than a “multidimensional” view of the problem. Historically, the general rule of thumb has been to look for other contexts where the same problem or phenomenon appears, and substantiate claims by evaluating across multiple tasks, datasets, distributions, or domains. - Superficial treatment of prior work: critical benchmarks, models, and related work omitted. Claims introduced in prior work repackaged in contemporary context without notable contribution: I expected to see either novel problem framing, or meaningful scope expansion. - Relatedly: hallucinated bibliographies (??!) This one is new to me, and a solid reason for a desk reject. The deluge of publications in recent years seems responsible for this trend, but shouldn’t excuse it. - Weak defensibility: design choices presented without justification via measurable improvements; critical baselines and ablations missing; architectural modifications appear arbitrary — and interestingly, can often be derived from existing methods with some linear algebra magic. This one may be evergreen, though the LLM era has really exacerbated the issue. - Chaotic paper structure: I may be biased because I was academically “raised” with guidelines drilled into my head on how to write a good paper — make contributions clear, substantiate every single one of them with experimentation, clearly differentiate from prior work, introduce strong baselines and evaluation frameworks, discuss failure modes, suggest follow-up directions for future work, etc etc Many colleagues have pointed this out already, but echoing it here; research has undergone a paradigm shift.

Goodluck everyone with the SIGGRAPH deadline today! If you're waiting for CVPR reviews, please note the following from the FAQ (s2026.siggraph.org/technical-pape…)

#SIGGRAPH2025 workshop on AI for Creative Visual Content Generation, Editing and Understanding has a full house now! @siggraph @cveu_workshop

in vancouver for @siggraph, let's meet?! on thursday (14) i'll be presenting at the @cveu_workshop my keynote is going to be "Generative AI as a medium of expression: beyond AI as a tool"

@multimodalart @siggraph Thanks for your contribution! Looking forward to your talk!

If you'd like to join our lunch gathering @ SIGGRAPH Here is the registration link forms.gle/nvtkJcAetzbw2c… Only successful registrations will be notified and invited through email about the location, and can come to the lunch.

You cannot miss this AIGC technical workshop in #SIGGRAPH2025 Thursday, August 14, 2025 9:00am - 12:15pm PDT, Vancouver, Canada (Room 11, East Building at Siggraph) Come to discuss with experienced professors and researchers on how to create eye-catching content and touching stories with AI @siggraph @VisLab_HKUST @magrawala @StanfordHCI @Adobe @huggingface @Snap @Snapchat cveu.github.io

Prof. Huamin Qu is the founding dean of AIS at HKUST, leads the VisLab @VisLab_HKUST, and is an award‑winning pioneer in explainable AI and creative visualization (IEEE VGTC, AI 2000, IBM Faculty Award...) and was inducted into the IEEE Visualization Academy in 2020. 🌟

Excited to announce that Prof. Huamin Qu (@HKUST) will give a talk at our #SIGGRAPH2025 workshop! 🎙️ Title: Human-AI Collaboration for Storytelling From data storytelling to screenwriting, explore how humans & AI co-create in the age of generative tech. #AI #HCI #Storytelling

Glad to have @magrawala as one of our keynote speakers at #SIGGRAPH2025 ! 🥳🥳He's going to give an exciting talk titled 'Making is Decision Making'. Check cveu.github.io for more updates and join discord.gg/Rgt6MBMT for live discussions with AI creators! 🍻

Ep2 is live. Guests: Chris Williamson & Jacky Zheng from Weta Workshop Truth bomb on AI in film: “I fear acceleration kills the LOVE of craft.” - Chris “Using AI? You’re in a constant BATTLE” - Jacky Will personalized AI stories ERASE water-cooler moments?

Ep1 live! Special guest: Tony Ngai, Asia Pacific Governor of SMPTE, shares how AI is transforming filmmaking, from workflows to the challenges of control. Advice to filmmakers: "Storytelling is your golden ticket OUT of film” What excites you most about AI in storytelling?

@CVPR #CVPR2025 #CVEU

The last session of CVEU today features a panel discussion with Google researcher @natanielruizg and @dutran , @MattNiessner , TUM professor and cofounder of @synthesiaIO and @SpAItial_AI, famous artist @_joelsimon ! Room 207 #CVPR25 Don’t miss it!

GenAI for games?! Welcome our speaker Nataniel Ruiz from Google! He is presenting generative infinite games: toward unbounded interactive worlds! Super exciting! Thanks @natanielruizg #CVPR25 #CVPR

Prof Matthias Niessner is now giving about 3D generation at the CVEU workshop in room 207! @MattNiessner #CVPR #CVPR25