The Release History of Juggernaut XL

Juggernaut XL did not arrive as one polished checkpoint. It evolved through a fast-moving series of SDXL releases, merge experiments, captioning upgrades, and later RunDiffusion collaborations. If you only know the model through v9, XI, or Ragnarok, the short version is this: Juggernaut started as a straight SDXL finetune in August 2023, became more photo-focused across the v6-v9 period, then shifted toward better captioning and stronger prompt response in X, XI, and the later family.

This timeline is based on the two internal reference documents in this repo plus public model pages and APIs. Where public records are incomplete, this article marks the gap instead of pretending the date is certain.

What Juggernaut XL Was Trying to Do

From the beginning, Juggernaut XL positioned itself as a practical SDXL workhorse rather than a niche art model. The recurring goals stayed fairly consistent:

deliver strong photorealistic output without heavy prompt ceremony
keep cinematic and editorial lighting attractive
improve anatomy, especially in the hands-and-feet problem area
remain usable across portraits, product shots, interiors, wildlife, and lifestyle imagery

That consistency is why the release history matters. Each version did not replace the identity of the model. It usually refined the same goal from a different angle: more steps, more side training, better photo merges, or better captions.

Phase One: The Fast Early Releases

The public Civitai API gives a surprisingly clear early timeline.

Version 1

Version 1 was published on August 22, 2023. Its public description is minimal, essentially just noting 220k Steps, but it is the first confirmed public Juggernaut XL release in the series.

Version 2

Version 2 followed on August 30, 2023. The published notes say the base model received 50k more steps, plus a second dataset trained with help from Dreamlook.AI for 58k steps. That matters because it established a pattern the model would revisit later: Juggernaut was not just being trained harder, it was being broadened through additional side data and merges.

Version 3

Version 3 was published on September 5, 2023. The public note says it added 100k more steps, started introducing different body types, and gave a minor whole-model quality upgrade. That sounds small, but it shows the project had already moved from generic realism toward specific output control.

Version 4 and 4.5

Version 4 (NSFW) appeared on September 13, 2023, followed by Version 4.5 on September 15, 2023. The 4.5 note describes it as a major fix for v4, including merges from older checkpoints and version 3. In other words, even this early, Juggernaut was already behaving like an iterative checkpoint family rather than a single linear model.

Version 5

Version 5 was published on September 21, 2023. Public notes point to a new side set added with Dreamlook.AI support. That is a small line in the changelog, but it reinforces the main pattern of the era: Juggernaut kept growing through side training, not only through one monolithic finetune.

Phase Two: The RunDiffusion Era Accelerates

The next phase is where Juggernaut XL became much easier to recognize in today's ecosystem. Public changelogs start connecting the model to RunDiffusion's photo-focused work and to a more explicit tuning philosophy.

Version 6 + RunDiffusion

Version 6 + RunDiffusion was published on October 25, 2023. Public notes describe:

Juggernaut 5.5 with another 200k steps
a Dreamlook.AI side model with 570k steps
integration of an unpublished RunDiffusion photo-real model
a baked-in VAE

This is one of the first releases where Juggernaut stops reading like a single checkpoint and starts reading like a production recipe.

Version 7 + RunDiffusion

V 7 + RunDiffusion landed on November 27, 2023. The public release note says the RunDiffusion photo model ratio was reduced and a Cinematic SideSet with 120k steps was added. The result was a more contrasty, more cinematic image profile. That is a useful clue if you compare outputs from the late-2023 family: the model was no longer only chasing realism, it was balancing realism against a stronger visual signature.

Version 8 + RunDiffusion

V 8 + RunDiffusion was published on January 6, 2024. The public note is short but unusually clear: it was a general quality update with focus on hands, feet, skin details, and photographic output.

This is important context for later versions. By v8, the author was already confronting the problem that every new training push improved some areas while hurting others.

Version 9 + RunDiffusion Photo 2

The standard V9 + RunDiffusionPhoto 2 release was published on February 18, 2024 on Civitai. The Hugging Face model repository for RunDiffusion/Juggernaut-XL-v9 was also created on February 18, 2024, which lines up with the public release window.

The v9 README is unusually detailed. It says:

the author only did basic training for skin details, lighting, and overall contrast
the biggest change came from RunDiffusion Photo Model v2
the model was delayed because further training had become harder without quality tradeoffs
this pressure is what pushed the author toward a full reboot plan for v10

That makes v9 a turning point. It is not just “another update.” It is the moment the public documentation admits that incremental training was reaching diminishing returns.

Version 9 Lightning

There is a second v9-era release that often gets mixed into the history: V9+RDPhoto2-Lightning_4S, published on February 23, 2024.

This is not the same thing as the standard v9 checkpoint. It is a later Lightning-oriented derivative for fast generation. When articles blur those two dates together, readers end up misunderstanding both the release timeline and the recommended settings.

Phase Three: Reboot Language and Post-v9 Maturity

After v9, Juggernaut XL moves into a more mature stage of its history. The public discussion shifts from “another merge update” to “how do we rebuild the series so it keeps improving.”

Juggernaut X

Jugg_X_by RunDiffusion was published on April 19, 2024. In the v9 README, v10 had already been framed as a reboot direction with re-captioning and better training foundations. By the time X ships, the branding has moved beyond plain numbers and toward named major releases.

Juggernaut XI

Jugg_XI_by_RunDiffusion was published on August 29, 2024. The public description says the whole Juggernaut 9 dataset was recaptioned and trained for XI across 15k images. That is one of the clearest signals in the entire model history: XI is not mainly about one clever merge ratio. It is about rebuilding data quality and prompt adherence through better captions.

Juggernaut XII

The exact standalone public release date for XII is less obvious than the dates for v1-v11, because the most accessible public RunDiffusion materials discuss XI and XII together rather than presenting a clean separate launch page with the same changelog precision as Civitai.

What can be stated safely is this:

by August 29, 2024, RunDiffusion was publicly publishing a prompt guide for Juggernaut XI and XII
RunDiffusion's public product pages describe XII as the more expressive, more cinematic sibling to XI
later documentation says Ragnarok used version 12 as its photorealism base

So XII clearly belongs to the later 2024 generation, even if its standalone release date is not as easy to source as the others.

Ragnarok

Ragnarok_by_RunDiffusion was published on May 7, 2025. Its public description says it used version 12 as the base to bring the model back toward photorealism, then merged in two NSFW-trained components:

one SDXL NSFW set at 0.15
one Lustify-based NSFW set at 0.10

The release note frames Ragnarok as a possible farewell edition for the project. It also makes a useful philosophical point: Juggernaut should be seen as a strong SDXL checkpoint in a wider pipeline, not as a model that magically beats every newer architecture in every category.

The Full Timeline at a Glance

Here is the compact version of the public release history:

August 22, 2023: Version 1
August 30, 2023: Version 2
September 5, 2023: Version 3
September 13, 2023: Version 4 (NSFW)
September 15, 2023: Version 4.5
September 21, 2023: Version 5
October 25, 2023: Version 6 + RunDiffusion
November 27, 2023: Version 7 + RunDiffusion
January 6, 2024: Version 8 + RunDiffusion
February 18, 2024: Version 9 + RunDiffusion Photo 2
February 23, 2024: V9 Lightning 4-step derivative
April 19, 2024: Juggernaut X
August 29, 2024: Juggernaut XI
By late 2024: Juggernaut XII is publicly documented in RunDiffusion materials
May 7, 2025: Ragnarok

Why This History Still Matters

Juggernaut XL remains interesting because its history is unusually transparent. Public release notes show the exact kinds of pressure that many SDXL model lines ran into:

more training does not automatically mean better output
photo quality and anatomy fixes can compete with other strengths
dataset curation and caption quality eventually matter more than one more merge
speed variants like Lightning can confuse a release history if you do not separate them from the base checkpoint

If you are evaluating Juggernaut today, the biggest lesson is not that every later version is “strictly better.” It is that each phase solved a different problem. Early versions established the look. The RunDiffusion era strengthened photographic output. XI pushed prompt fidelity through recaptioning. Ragnarok closed the loop by pulling the family back toward photorealism with a specific v12-based formula.