No tags are required, but read below to get the most out of this LoRA!

Recommended Tags: from_behind, television, watching_television, side-by-side, facing_away,

Lora strength Suggestion: 0.5 to 1.0

Sampling Method: Eular, DPM++ 2M, @20 to 30 steps

Purpose: The point of this LoRA was to describe a scene where a couple is watching television in their living room. As we call it "Netflix 'n' chill." Though that term became synonymous with the act of engaging in sex. Older folks and younger people alike still take it to simply mean: watch streaming movies on television and relax. However, this LoRA was exceedingly complicated to make, and instead of my usual "one LoRA a day" quota, it took me 2 days of training both on my main PC, spare PC (both using Kohya) and Civitai's online trainer. I would highly advise you read the following to get the most out of this, as describing this scene in general without a LoRA is also fairly difficult even using regional prompter or Forge Couple. Disclaimer: This LoRA has nothing to do with Netflix. Apologies if that was your intent with this LoRA.

Suggestions:

-Always describe your couch color, and tag the shot as "from_behind."
-Turn up the strength depending on your scene. At 1.0, the couple or people are always watching TV and the scene is composed to include the shot of a TV and anything else in the room (such as "plants, TV_stand, rug, table, lamp, window, curtains, etc."
-At 0.5, the couple can freely engage in actual "Netflix'n'chillin." I have included sex scenes in the gallery so you can copy and play with your own prompts

-At the end of each tag, include what you would like the television to display. Some tags don't work well, others are spontaneous. Sometimes descriptions of the room end up displayed on the TV. Some tags that work: pro wrestling, various pokemon (though most of the time it just shows a very generic Pokemon that looks like all of them), various Disney characters, monsters, explosions, cars, etc. Play around with this and if it doesn't display what you want, go ahead and cheat and inpaint it.

-I'll describe complications at the end of this description.

EXAMPLES)

8K,absurdres,very aesthetic,masterpiece,highly_detailed,best_quality,newest,t-shirt,dress,pants,shorts,brown hair,black hair,window,curtains,door,realistic,side-by-side,sitting,speakers,plants,couple,1boy,1girl,1girl,face,brown hair,from behind,television,bear,teddy bear,smile,facing away,couch,

Steps: 30, Sampler: DPM++ 2M, Schedule type: Karras, CFG scale: 7, Seed: 418322496, Size: 896x1152 *LoRA strength: 0.9

8K,absurdres,very aesthetic,masterpiece,highly_detailed,best_quality,newest,t-shirt,dress,pants,shorts,brown hair,black hair,window,curtains,door,side-by-side,sitting,speakers,plants,couple,1boy,1girl,1girl,face,brown hair,from behind,game controller,playing games,television,from above,monster,final fantasy,fighting,

Steps: 30, Sampler: DPM++ 2M, Schedule type: Karras, CFG scale: 7, Seed: 3397138112, Size: 896x1152 *LoRA strength: 0.8

8K,absurdres,very aesthetic,masterpiece,highly_detailed,best_quality,newest,multiple people,blonde,t-shirt,dress,pants,shorts,brown hair,black hair,1boy,2girls,window,curtains,door,speakers,television,television,side-by-side,from behind,brown_couch,sitting,pro wrestling,mask,wrestling ring,arena,

Steps: 30, Sampler: Euler, Schedule type: Automatic, CFG scale: 7, Seed: 993932818, Size: 1152x896 *LoRA strength: 1.0

Complications/Issues)

i) This LoRA does some weird things. It's very temperamental. Sometimes the couch/sofa doesn't look right, sometimes it'll place the couch facing the viewer anyway, despite tagging otherwise. This is just one of those LoRAs you have to play with to get right for your checkpoint of preference. For example:

ii) For spicier scenes, you have to turn down the dials on the LoRA. This means to suppress it's strength from 1 to around 0.5. At 1.0 there's literally nothing you can do with the couple or people. All you get is the back of they heads and some clothes options. At around 0.7 is when they can start doing some things, but 0.5 preserves some functions of the LoRA while allowing some wiggle room to do things. See the gallery for the spicier prompts that are NSFW.

iii) SIZE MATTERS! Yes, that's right, I said it. Not just in the physical traits you inherited department, but also in your picture size. Landscape pictures will produce different pictures versus portrait. Also, 768x1344 tends to dump people on top of each other.

iv) 20% of the dataset included 2 additional angles - one from the front, depicting the couple or people from behind the TV, and one from the side. For whatever reason, despite tagging "behind_television" and "from_side," they don't work on outputs. Even dialing the strength up and down. To get these prompts to work, I actually removed the LoRA altogether and got decent results but still no "behind_television."

v) The original activation tag was "tvwatch." I included this in the original Kohya WD14 tag, and reintroduced it in Civitai Training mode. The tag doesn't work at all and was supposed to unify the "behind_television" and "from side" angles. You can try this out when generating your prompts but I found that it serves no purpose.

...and then this pic which helps to explain my frustration with putting together this LoRA:

PLEASE SEE THE GALLERY PICTURES FOR MORE INFORMATION TO AID YOUR PROMPTS!

Background information no one cares about:

Since I began experimenting with Stable Diffusion, it's always been my desire to "tell a story." One of the most depicted compositions in film, tv, illustration and photography is the scene where a couple are sitting on a couch, watching television. The scene always involves the back of their heads, a couch, and focus on the TV they are viewing. It could be a romantic movie, a news flash, an action, sports, etc. For obvious and not so obvious reasons, this kind of scene is terribly complicated to prompt out. 9 times out of 10 I've had to use img2img to get things right. I've wanted to make this LoRA since the beginning of my journey in Stable Diffusion and well, it's finally here. I really hope to make improvements on this model, then perhaps I can focus on other LoRAs including finishing Disneyland, doing something with ghosts, and more NSFW ones. The NSFW ones pay for the passion projects, YKWIS?

Search AI Models & Addons

very few results